idnits 2.17.1 draft-taddei-smart-cless-introduction-03.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (July 13, 2020) is 1382 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Unused Reference: 'EPTAXONOMY' is defined on line 1696, but no explicit reference was found in the text == Outdated reference: A later version (-02) exists of draft-mcfadden-smart-endpoint-taxonomy-for-cless-00 Summary: 0 errors (**), 0 flaws (~~), 3 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 IETF A. Taddei 3 Internet-Draft Broadcom 4 Intended status: Informational C. Wueest 5 Expires: January 14, 2021 Acronis 6 K. Roundy 7 Norton Lifelock 8 D. Lazanski 9 Last Press Label 10 July 13, 2020 12 Capabilities and Limitations of an Endpoint-only Security Solution 13 draft-taddei-smart-cless-introduction-03 15 Abstract 17 In the context of existing, proposed and newly published protocols, 18 this draft RFC is to establish the capabilities and limitations of 19 endpoint-only security solutions and explore benefits and 20 alternatives to mitigate those limits with the support of real case 21 studies. 23 Status of This Memo 25 This Internet-Draft is submitted in full conformance with the 26 provisions of BCP 78 and BCP 79. 28 Internet-Drafts are working documents of the Internet Engineering 29 Task Force (IETF). Note that other groups may also distribute 30 working documents as Internet-Drafts. The list of current Internet- 31 Drafts is at https://datatracker.ietf.org/drafts/current/. 33 Internet-Drafts are draft documents valid for a maximum of six months 34 and may be updated, replaced, or obsoleted by other documents at any 35 time. It is inappropriate to use Internet-Drafts as reference 36 material or to cite them other than as "work in progress." 38 This Internet-Draft will expire on January 14, 2021. 40 Copyright Notice 42 Copyright (c) 2020 IETF Trust and the persons identified as the 43 document authors. All rights reserved. 45 This document is subject to BCP 78 and the IETF Trust's Legal 46 Provisions Relating to IETF Documents 47 (https://trustee.ietf.org/license-info) in effect on the date of 48 publication of this document. Please review these documents 49 carefully, as they describe your rights and restrictions with respect 50 to this document. Code Components extracted from this document must 51 include Simplified BSD License text as described in Section 4.e of 52 the Trust Legal Provisions and are provided without warranty as 53 described in the Simplified BSD License. 55 Table of Contents 57 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 58 2. Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . 4 59 3. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 6 60 4. Disclaimer . . . . . . . . . . . . . . . . . . . . . . . . . 6 61 5. Endpoints: definitions, models and scope . . . . . . . . . . 6 62 5.1. Internal representation of an endpoint . . . . . . . . . 7 63 5.2. Endpoints modeled in an end-to-end context . . . . . . . 8 64 6. Threat Landscape . . . . . . . . . . . . . . . . . . . . . . 9 65 7. Endpoint Security Capabilities . . . . . . . . . . . . . . . 11 66 8. What would be a perfect endpoint security solution? . . . . . 14 67 9. The defence-in-depth principle . . . . . . . . . . . . . . . 15 68 10. Endpoint Security Limits . . . . . . . . . . . . . . . . . . 17 69 10.1. No possibility to put an endpoint security add-on on the 70 UE . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 71 10.1.1. Not receiving any updates or functioning patches . . 19 72 10.1.2. Mirai IoT bot . . . . . . . . . . . . . . . . . . . 19 73 10.2. Endpoints may not see the malware on the endpoint . . . 21 74 10.2.1. LoJax UEFI rootkit . . . . . . . . . . . . . . . . . 21 75 10.2.2. SGX Malware . . . . . . . . . . . . . . . . . . . . 22 76 10.2.3. AMT Takeover . . . . . . . . . . . . . . . . . . . . 22 77 10.2.4. AMT case study (anonymised) . . . . . . . . . . . . 23 78 10.2.5. Users bypass the endpoint security . . . . . . . . . 24 79 10.3. Endpoints may miss information leakage attacks . . . . . 24 80 10.3.1. Meltdown/Specter . . . . . . . . . . . . . . . . . . 24 81 10.3.2. Network daemon exploits . . . . . . . . . . . . . . 24 82 10.3.3. SQL injection attacks . . . . . . . . . . . . . . . 25 83 10.3.4. Low and slow data exfiltration . . . . . . . . . . . 25 84 10.4. Suboptimality and gray areas . . . . . . . . . . . . . . 26 85 10.4.1. Stolen credentials . . . . . . . . . . . . . . . . . 26 86 10.4.2. Zero Day Vulnerability . . . . . . . . . . . . . . . 27 87 10.4.3. Port scan over the network . . . . . . . . . . . . . 27 88 10.4.4. DDoS attacks . . . . . . . . . . . . . . . . . . . . 28 89 11. Learnings from production data . . . . . . . . . . . . . . . 29 90 11.1. Endpoint only incidents . . . . . . . . . . . . . . . . 30 91 11.2. Security incidents detected primarily by network 92 security products . . . . . . . . . . . . . . . . . . . 31 93 11.2.1. Unauthorized external vulnerability scans . . . . . 31 94 11.2.2. Unauthorized internal vulnerability scans . . . . . 32 95 11.2.3. Malware downloads resulting in exposed endpoints . . 32 96 11.2.4. Exploit kit infections . . . . . . . . . . . . . . . 32 97 11.2.5. Attacks against servers . . . . . . . . . . . . . . 33 98 12. Regulatory Considerations . . . . . . . . . . . . . . . . . . 34 99 12.1. IoT Security . . . . . . . . . . . . . . . . . . . . . . 34 100 12.2. Network infrastructure . . . . . . . . . . . . . . . . . 35 101 12.3. Auditing and Assessment . . . . . . . . . . . . . . . . 35 102 12.4. Privacy Considerations . . . . . . . . . . . . . . . . . 36 103 13. Human Rights Considerations . . . . . . . . . . . . . . . . . 36 104 14. Security Considerations . . . . . . . . . . . . . . . . . . . 36 105 15. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 36 106 16. Informative References . . . . . . . . . . . . . . . . . . . 36 107 Appendix A. Contributors . . . . . . . . . . . . . . . . . . . . 41 108 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 42 110 1. Introduction 112 This Internet Draft aims to be a reference to the designers of 113 protocols on the capabilities and limitations of security solutions 114 on endpoint devices against malware and other attacks. As security 115 is entering a new phase in the arms race between attackers and 116 defenders, with many technical, economic and regulatory changes, and 117 with a significant increase in major data breaches, it is a good 118 moment to propose a systematic review and update on what is an old 119 and constantly evolving problem: endpoint security. 121 With the above context in mind this document will focus on the 122 capabilities and limitations of an endpoint-only security solution. 124 We want to explore a number of questions: 126 o What endpoint models do we have? 128 o What is the threat landscape under consideration? 130 o Can we differentiate security and privacy threats? 132 o What are common endpoint security capabilities? 134 o What would be an ideal endpoint security solution? 136 o What are the limits to endpoint security? 138 o What is real production data telling us? 140 o What can defence-in-depth help us with? 142 o What are the economic considerations? 144 o What are the regulatory considerations and constraints? 145 o What are the human rights considerations? 147 Our goal with this review is to describe the benefits and limitations 148 of endpoint security in the real world, rather than in the abstract. 149 We aim to highlight security limitations that cannot be addressed by 150 endpoint solutions and to suggest how these may be mitigated with the 151 concept of a defence-in-depth approach, in order to increase the 152 resilience against attacks and data breaches. 154 2. Abbreviations 156 In this section we provide main abbreviations expansions 158 ABAC Attribute Based Access Control 160 AI Artificial Intelligence 162 AMT Active Management Technology 164 C&C Command and Control 166 CFI Control Flow Integrity 168 CFG Control Flow Guard 170 DDoS Distributed Denial of Service 172 DEP Data Execution Prevention 174 DGA Domain Generating Algorithms 176 DLP Data Loss Prevention 178 DMARC Domain-based Message Authentication, Reporting and Conformance 180 DoS Denial of Service 182 EE Execution Environment 184 EDR Endpoint Detection and Response 186 EPP Endpoint Protection Platform 188 FP False Positive 190 HIPS Host Intrusion Prevention System 192 ICD Integrated Cyber Defence 193 ICMP Internet Control Message Protocol 195 IDS Intrusion Detection System 197 IoT Internet of Things 199 IPS Intrusion Prevention System 201 ML Machine Learning 203 MSS Managed Security Services 205 MSSP Managed Security Services Provider 207 NIST National Institute of Standards and Technology 209 NX No Execute Bit 211 P2P Peer to Peer 213 RAP Reuse Attack Protector 215 RBAC Role Based Access Control 217 RDP Remote Desktop Protocol 219 ROP Return Oriented Programming 221 SANS System Administration, Networking, and Security 223 SGX Software Guard eXtensions 225 SSH Secure SHell 227 UE User Equipment 229 UEFI Unified Extensible Firmware Interface 231 UX User Experience 233 VM Virtual Machine 235 XSS Cross Site Scripting 237 3. Definitions 239 In this section we provide definitions that are marked 241 o (L) Local to this document 243 o (G REFERENCE) Global and then will be preceded by a reference 245 DoS (L) Literally a Denial of Service. Not to be confused with a 246 Network DoS or DDoS. 248 Endpoint security capabilities (L) How to protect the endpoint with 249 three different aspects of protection: 251 o Prevention - The attack doesn't succeed by intrinsic or explicit 252 security capabilities. 254 o Detection - The attack is happening or has happened and is 255 recorded and/or signalled to another component for action. 257 o Mitigation - Once detected, the attack can be halted or its 258 effects can at least be reduced or reversed. 260 System (L) A system is a heterogeneous set of any IT capabilities 261 including hardware, software, endpoints (including IoT), networks, 262 data centers and platforms with no assumptions on deployment form 263 factor (physical, virtual, microservices), deployment scenario, 264 geographic distribution, or dispersion. 266 User Equipment (G ITU-T H.360) Equipment under the control of an End 267 User 269 4. Disclaimer 271 This document is a first draft and is incomplete on purpose. Indeed 272 there are several areas where there are different ways to develop 273 this draft and the authors are seeking for feedback and extended 274 collaboration. This is to be noted too, that this is the first draft 275 RFC for the authors and contributors, so, coaching and help will be 276 appreciated. Overall, 'a bon entendeur, salut'. 278 Comments are solicited and should be addressed to the authors. 280 5. Endpoints: definitions, models and scope 282 Endpoints are the origin and destination for a communication between 283 parties. This encompasses User Equipment (UE) and the Host at the 284 other end of the communication. Whilst it is recognized that these 285 two ends of the communication may represent a vast amount of diverse 286 endpoints, this document will set here a requirement for a uniform 287 way to describe the endpoints in order to work from an equally 288 uniform representation of what is called the Attack Surface. In the 289 same spirit as the IETF TEEP Working Group generalized its work, see 290 [TEEP], this document will rely on another document identified as 291 [I-D.draft-mcfadden-smart-endpoint-taxonomy-for-cless-00] in order to 292 represent the taxonomy of endpoints. 294 For example: 296 o The following would be considered as UEs: a smartphone, a smart 297 device, any IoT device, a laptop, a desktop, a workstation, etc. 299 o Hosts represent too, physical servers, virtual servers/machines, 300 etc. 302 We require a framework in order to define and model the endpoint 303 itself and the position of the endpoint in the network. In this 304 initial analysis we focus on endpoints that are User Equipment (UE) 305 rather than on hosts. In the future, with the help of 306 [I-D.draft-mcfadden-smart-endpoint-taxonomy-for-cless-00] we hope to 307 balance and unify the model. 309 In addition, we need two models for the endpoint, internally and in 310 an end-to-end context within the network. With this approach we 311 expect both models to help us cover all the attack surface and the 312 threat landscape and therefore help us list the capabilities and 313 limitations for endpoint security. 315 Indeed, this will help us understand point attacks versus composite 316 attacks within context, and, accordingly, understand holistically the 317 capabilities and the limitations of endpoint security. For example 318 to differentiate when only an application on the end point is 319 affected. 321 5.1. Internal representation of an endpoint 323 This section interfaces here with 324 [I-D.draft-mcfadden-smart-endpoint-taxonomy-for-cless-00] which 325 starts from the below internal representation of an endpoint which 326 could be generalized by the simple diagram below: 328 +----------------------------+ 329 | Application | 330 +----------------------------+ 331 | OS / Execution Environment | 332 +----------------------------+ 333 | Hardware | 334 +----------------------------+ 336 Today there are many combinations of Hardware, OS/EE pairing and 337 Application layers, offering the user a vast set of features with a 338 wide spectrum of capabilities. 340 Furthermore we can consider that an application running on a UE or a 341 host is an endpoint too, so we have multiple ways to read the above 342 diagram. 344 In essence we want to consider here endpoints including those which 345 have a variance in electrical power, computational power, memory, 346 disk, network interfaces, size, ownership, connectics, etc. and 347 therefore why we rely on 348 [I-D.draft-mcfadden-smart-endpoint-taxonomy-for-cless-00]. 350 5.2. Endpoints modeled in an end-to-end context 352 A representation of endpoints in an end-to-end context could look 353 like the following diagram: 355 +-------+ +---------------------+---------+ 356 | Human | <- (1) -> | Digital Persona | Application | <- (2) -> 357 +-------+ +-----------+-------------------+ 358 | User Equipment | 359 +-------------------------------+ 361 +----------------+ +----------------+ 362 <- (2) -> | Network | <- (3) -> | Platform/Hosts | 363 | Infrastructure | +----------------+ 364 +----------------+ 366 1. Humans have a user experience (UX) with the UE, starting with an 367 explicit or implicit Digital Persona, engaging with an 368 application 370 2. The application will have sessions through a large Network 371 Infrastructure where we do not assume anything of the 372 infrastructure (could be landlines, mobile networks, satellites, 373 etc.) and those sessions reach 375 3. a Platform consisting of many Hosts either physical or virtual 376 and it ensures a large part of the end-to-end user experience. 378 In this end-to-end model we see that many other systems may have 379 interactions with the UE: the human, the UX, the digital persona, the 380 sessions, the intermediate network infrastructure, and the hosts and 381 application at the destination. 383 If we now look at security aspects of the above models, the threat 384 landscape is very large and the attack surface will cover all the 385 components and interactions at any level. 387 6. Threat Landscape 389 (Editor's note: this section will require a significant amount of 390 future development.) 392 Given the vast number of combinations that the above generic modeling 393 offers us, defining a threat landscape should be done carefully and 394 will require a systematic methodology. 396 Therefore this entire section will be developed through future 397 iterations of the document, in this initial version we will start 398 structuring an approach and then adjust this based on feedback. 400 There is no doubt that we want to cover typical known attacks such 401 as: 403 o Malware (Trojans, viruses, backdoors, bots, etc.) 405 o Adware and spyware 407 o Exploits 409 o Phishing 411 o Script based attacks 413 o Ransomware, local Denial of Service (DoS) attacks 415 o Denial of Service (DoS) attacks 417 o Malicious removable storage devices (USB) 418 o In memory attacks 420 o Rootkits and firmware attacks 422 o Scams and online fraud 424 o System abuse (staging/proxying) 426 o etc. 428 To illustrate the difficulty to define a good threat landscape, when 429 it comes to cryptojacking and coinmining that were on the rise, in 430 which category do they fall: malware? DoS? system abuse? or a 431 category on its own? 433 This is why we wanted to conduct a thorough gap analysis using 434 existing definitions and frameworks, but we couldn't find an existing 435 comprehensive and recognized taxonomy dedicated to the threat 436 landscape on endpoints. We found however different models in this 437 field, and have considered two. We are open to further suggestions. 439 Indeed both of the analysed frameworks contain threat landscape 440 descriptions: 442 o MITRE Common Attack Pattern Enumeration Classification (CAPEC). 443 See [CAPEC]. 445 o MITRE ATT&CK. See [ATTACK]. 447 These offer us interesting ways to assess the threat landscape: 449 o CAPEC offers a hierarchical view of attack patterns by domains 450 which can match some aspects of both of our above models, but we 451 will need to identify those attacks that fit exactly in our scope. 453 o ATT&CK offers a very straightforward categorized knowledge base of 454 attacks, but it concentrates on the entreprise attack chain, so we 455 will need to do some work to extract what we need. 457 We recognise however that these frameworks do not address all of the 458 threats that can affect the security of a system, for example they do 459 not cover; routing hijacking, flooding, selective blocking, 460 unauthorised modification of data sent to an endpoint, etc. Further 461 work to define categories of threats is therefore required. 463 As a further example, phishing should be included as an attack, but 464 whilst this is indeed an attack that will materialize on a device 465 through an application (email, webmail, etc.), the real target of 466 this attack is not the device, but the human behind the digital 467 persona. 469 Having a methodology of assessment is necessary here, because it will 470 help decide what is in scope vs. out of scope. 472 We are aware that once a method and the categories are fully defined 473 in this section, it will force a review of all the following sections 474 in the document. Whilst remapping will be necessary, it should not 475 drastically change the draft. 477 7. Endpoint Security Capabilities 479 In this section we try to define some endpoint security capabilities 480 (Editor's note: this section will require future development.) 482 In this version of the document we will start by developing a 483 framework to categorize and position endpoint security capabilities 484 with the goal of defining what an ideal endpoint security capability 485 would look like. 487 By endpoint security capabilities we mean how to protect the endpoint 488 against attacks. Protection has many meanings, we want to 489 distinguish three different aspects of protection: 491 o Prevention - The attack doesn't succeed by intrinsic or explicit 492 security capabilities. 494 o Detection - The attack is happening or has happened and is 495 recorded and/or signalled to another component for action. 497 o Mitigation - Once detected, the attack can be halted or its 498 effects can at least be reduced or reversed. 500 For example, prevention methods include keeping the software updated 501 and patching vulnerabilities, implementing measures to authenticate 502 the provenance of incoming data to stop the delivery of malicious 503 content, or choosing strong passwords. Detection methods include 504 inspecting logs or network traffic. Mitigation could include 505 deploying backups to recover from an attack with minimal disruption. 507 Our intention however is not just to consider each endpoint security 508 capability separately, but also the overall endpoint security 509 holistically with all its interdependencies. Indeed, we defined a 510 simple endpoint, but each layer may or may not have a certain 511 spectrum of intrinsic capabilities and there may be multiple ways to 512 provide add-on and third-party endpoint security capabilities, 513 allowing complex interactions between all of these components. 515 We define two different aspects of endpoint security capabilities and 516 their subdivisions as: 518 o (A) Intrinsic security capability can be built-into each of the 519 endpoint model layers 521 * (1) Hardware 523 * (2) OS/EE 525 * (3) Application 527 o (B) Add-on security capability can be 529 * (4) a component of the hardware 531 * (5) a component of the OS/EE 533 * (6) an application by itself 535 In (A) we relate to a 'security by design' intention of the 536 developers and they will intrinsically offer a security model and 537 security capabilities as part of their design. A typical example of 538 this is the authorization model. 540 In (B) a 3rd party is offering an additional security component which 541 was not necessarily considered when the Hardware, OS/EE or 542 Application were designed. 544 In the future we will review all the main categories of security 545 capabilities that are known to date and assess security capability 546 enablers like Artificial Intelligence (AI) and Machine Learning (ML). 547 For each category we will try to give a review on how effective the 548 capability is in securing the system. 550 With regard to (6), there are many available options for add-on 551 security capabilities offered by third-parties as applications on a 552 commercial or open-source basis. Gartner (see [GARTNERREPORT]) 553 highlights the evolution of endpoint security towards two directions 554 as shown in [EPPEDR], [EPPSECURITY], [EPPGUIDE]. 556 o Endpoint Protection Platform (EPP) as an integrated security 557 solution designed to detect and block threats at the device level. 559 o Endpoint Detection and Response (EDR) as a combination of next 560 generation tools to provide anomaly detection and alerting, 561 forensic analysis and endpoint remediation capabilities. 563 Among the security capabilities that we list, the endpoint can 564 perform the following: 566 o Intrinsic 568 * Software updates / patching 570 * Access Control (RBAC, ABAC, etc.) 572 * Authentication 574 * Authorization 576 * Detailed event logging 578 o Execution protection 580 * Exploit mitigation (file/memory) 582 * Tamper protection 584 * Whitelisting filter by signatures, signed code or other means 586 * System hardening and lockdown (HIPS, trusted boot, etc.) 588 o Malware protection 590 * Scanning - on access/on write/scheduled/quick scan (file/ 591 memory) 593 * Reputation-based blocking on files or by ML 595 * Behavior-based detection - (heuristic based/ML) 597 * Rootkit and firmware detection 599 * Threat intelligence based detection (cloud-based/on premise) 601 * Static detection - generic, by emulation, by ML, by signature 603 o Attack/Exploit/Application Protection 605 * Application protection (browser, messaging clients, social 606 media, etc.) 608 + Disinformation Protection (anti-phishing, fake news, anti- 609 spam, etc.) 611 + Detection of unintended link location (URL blocklist, etc.) 613 + Memory exploit mitigation, e.g. browsers 615 * Network Protection (local firewall, IDS, IPS and local proxy) 616 inbound and outbound 618 * Detection of network manipulation (ARP, DNS, etc.) 620 * Data Loss Prevention and exfiltration detection (incl. covert 621 channels) 623 8. What would be a perfect endpoint security solution? 625 With all the above knowledge, let's consider what we could expect 626 from a perfect endpoint security 'system'. It would: 628 o find instantly accurate reputation for any file before it gets 629 executed and block it if needed. 631 o monitor any behavior on the endpoint, including inbound and 632 outbound network traffic, learn and identify normal behavior and 633 detect and block malicious actions, even if the attack is misusing 634 legitimate clean system tools or hiding with a rootkit. 636 o patch instantly across all devices/systems/OSes, including virtual 637 patching, meaning you can patch or shield an application even 638 before an official patch is released. 640 o exploit protection methods for all processes where applicable, 641 e.g. no execute bit (NX), data execution prevention (DEP), address 642 space layout randomization (ASLR), Control Flow Integrity Guard 643 (CFI/CFG), stack canaries, shadow stack, reuse attack protection 644 (RAP), etc. all of which are methods, which make it very difficult 645 to successfully run any exploit, even for zero day 646 vulnerabilities. 648 o detect attempts to re-route data to addresses other than those 649 which the user intended, e.g. detect incorrectly served DNS 650 entries, TLS connections to sites with invalid certificates, data 651 that is being proxied without explicit user consent, etc. 653 o have an emulator/sandbox/micro virtualization to execute code and 654 analyse the outcome and perform a roll back of all actions if 655 needed, e.g. for ransomware. 657 o allow the endpoint to communicate with the other endpoints in the 658 local network and globally, to learn from 'the crowd' and 659 dynamically update rules based on its findings. 661 o be in constant sync with all other endpoints deployed on a network 662 and other security solutions, run on any OS, with no delay 663 (including offline modes and on legacy systems). 665 o run from the OS/EE when possible. 667 o run as one of the first process on the OS/EE and protect itself 668 from any form of unwanted tampering. 670 o offers a reliable logging that can't be tampered with, even in the 671 event of system compromise. 673 o receive updates instantly from a trusted central entity. 675 9. The defence-in-depth principle 677 In this section we give a high level view of what we mean by 678 'defence-in-depth'. 680 Whilst endpoint security systems have good capabilities, sometimes it 681 is debatable and perhaps suboptimal to let the endpoint run the 682 capability alone or at all. It is generally considered good security 683 practice to adopt a defence-in-depth approach (see [USCERT]). The 684 Open Web Application Security Project group (OWASP) describes the 685 concept as follows: "The principle of defense-in-depth is that 686 layered security mechanisms increase security of the system as a 687 whole. If an attack causes one security mechanism to fail, other 688 mechanisms may still provide the necessary security to protect the 689 system." (see [OWASP]) 691 Indeed there are many other constituencies as per our end-to-end 692 model that can participate in the defence process: The network, the 693 infrastructure itself, the platform, the human, the user experience 694 and in a hybrid of an on premise and cloud approach, an Integrated 695 Cyber Defence (ICD) of the entire chain. 697 The simple idea behind the concept is that "every little helps". If 698 the endpoint is not 100% secure itself, the detection chance can 699 increase with additional security capabilities from other entities. 700 We acknowledge that there are some case where adding an additional 701 component to the system may degrade the overall security level by 702 introducing new weaknesses. 704 There are various reference article in the industry highlighting 705 limitations of endpoint only solutions. For example this quote here, 706 which talks about multi-tier solutions: "There are limitations with 707 any endpoint protection solution, however, that can limit protection 708 to only the client layer. There is also a need for security above 709 the client layer, as endpoint protection products cannot intercept 710 traffic. Vendors will often sell a multi-tiered solution that 711 enables a network appliance to assist the endpoint protection client 712 by intercepting traffic between the attacker and the infected client. 713 Vendors will also sell solutions that monitor and intercept traffic 714 on internal or external network segments to protect the enterprise 715 from these threats. A prime example of the limitations of endpoint 716 protection software is infection via a phishing attack." [ADAPTURE]. 718 Some sources point out that even the best solution might not get 719 deployed in the optimal way in a real world scenario as the 720 environment can be very complex: "While endpoint security has 721 improved significantly with the introduction of application 722 whitelisting and other technologies, our systems and devices are 723 simply too diverse and too interconnected to ensure that host 724 security can be deployed 100% ubiquitously and 100% effectively." 725 [NETTODAY] 727 On these grounds it is considered a good idea to follow a layered 728 approach when it comes to security. "In today's complex threat 729 environment, companies need to adopt a comprehensive, layered 730 approach to security, which is a challenging task in such as rapidly 731 evolving, crowded market." [HSTODAY] 733 It is important to comprehend the capabilities of endpoint security 734 solutions in this overall picture of the connected environment, which 735 includes other systems, networks and various protocols that are used 736 to interact with these entities. Understanding possible shortcomings 737 from single layered solutions can help counterbalance such weaknesses 738 in the architectural concept or the protocol design. 740 In order to quantify any potential benefits or limitations of the 741 various layered scenarios in regards to security a solid data set is 742 needed. This section requires statistics about proportions of 743 attacks that go undetected in various cases. We propose analysing 744 data for the following four cases: 746 o There is no security solution 748 o Security is only on the endpoint 750 o Security is only on the network 751 o Security is on both the endpoint and the network 753 However reconciling various statistics requires a lot of caution and 754 time, a methodology and consistent classification to avoid any 755 misinterpretation. 757 10. Endpoint Security Limits 759 The previous section defines an ideal endpoint security 'system', 760 however, from the real world, the expectation of what we can get from 761 an endpoint security solution will look more along the following 762 lines: 764 o may not be able to run at full capacity due to computational power 765 limits, battery life, performance, or policies (such as BYOD 766 restrictions in enterprise networks), etc. 768 o may not be able to run at full capacity as it slows down 769 performance too much. 771 o will miss some of the malware or attacks, regardless of detection 772 method used, like signatures, heuristics, machine learning (ML), 773 artificial intelligence (AI), etc. 775 o have some level of False Positives (FP). 777 o not monitoring or logging all activities on the system, e.g. due 778 to constraints of disk space or when a clean windows tool is being 779 triggered to do something malicious but the activity is not 780 logged. Such activity can be logged, but a decision needs to be 781 made if it's clean or not. 783 o have its own vulnerabilities or simple instabilities that could be 784 used to compromise the system. 786 o be tampered with by the user, e.g. disabled or reconfigured. 788 o be tampered with by the attacker, e.g. exceptions added or log 789 files wiped. 791 In the section below we review a number of these limitations through 792 real examples, step by step. Some limitations are absolute, and some 793 limitations result in a grey area or suboptimality for the solution. 795 10.1. No possibility to put an endpoint security add-on on the UE 797 UEs will vary a lot; by 2022, an estimated 29 billion devices will be 798 connected, with 18 billion of them related to IoT [ERICSSON]. Many 799 IoT products lack the capacity to install any endpoint security 800 capabilities, are unable to update the software, and it is not 801 possible to force the UE provider to improve or even offer an 802 intrinsic security capability. 804 We acknowledge that the numbers do vary significantly depending on 805 the source, for example: 807 o [STATISTA1] is showing the current trajectory of IoT devices from 808 25B to date to 40+B in 2022 and 75B in 2025. 810 o [ERICSSON] is more conservative and might requires an update, but 811 it was reaching 29B devices in 2022, with a nice breakdown between 812 device types and connectivity. 814 o [STATISTA2] is showing a breakdown by verticals and is even more 815 conservative than both of the above. 817 o [ENISA] it refers to a [GARTNERIOT] report from 2017 which sets a 818 trajectory to 20B devices by 2020. 820 In IoT we find UEs such as medical devices which are limited by 821 regulation, welding robots that can't be slowed down, smart light 822 bulbs which are limited by the processing power, etc. There are many 823 factors influencing whether endpoint security can be added to a UE: 825 o The UE is simply not powerful enough or the performance hit is too 826 high. 828 o Adding your own security will breach the warranty or will 829 invalidate a certification or a regulation (breach of validity). 831 o The UE needs to run in real-time and any delay introduced by a 832 security process might break the process. 834 o Some UEs are simply locked by design and the manufacturer does not 835 provide a security solution (e.g. smart TV, fitness tracker or 836 personal artificial assistants) see [CANDID1], [CANDID2]. 838 In the future, a possible research problem would be to find hard data 839 on the exact proportion of IoT devices that are unable to run any 840 endpoint security add-on or that have no intrinsic security built-in. 842 The other hidden dimension here is the economical aspect. Many 843 manufacturer are reluctant to invest in IoT device security, because 844 it can significantly increases the cost of their solution and there 845 is the perception that they will lose market shares, as customers are 846 not prepared to pay the extra cost for added security. 848 10.1.1. Not receiving any updates or functioning patches 850 The endpoint security system may lack a built-in capability to be 851 patched or it may be connected to a network that prevents the process 852 of downloading updates automatically. For example stand-alone 853 medical systems or industrial systems in isolated network segments 854 often do not have a communication channel to the Internet. 856 Even if security updates are received, they typically will only be 857 periodically updated; hence there will be a window of opportunity for 858 an attacker, between the time the attack is first used, and the time 859 the attack is discovered/patched and the patch is deployed. 861 In addition updates and patches may themselves be malicious by 862 mistake, or on purpose if not properly authenticated, or if the 863 source of the updates has malicious intent. This could be part of a 864 software update supply chain attack or an elaborate attacker breaking 865 the update process, as for example seen with the Flamer group (see 866 [FLAMER]). 868 A recent survey found that fewer than 10% of consumer IoT companies 869 follow vulnerability disclosure guidelines at all, which is regarded 870 as a basic first step in patching vulnerabilities (see 871 [IOTPATCHING]). This indicates that many IoT devices do not have a 872 defined update process or may not even create patches for most of the 873 vulnerabilities. 875 Furthermore some endpoints system may reach the end of their support 876 period and therefore no longer receive any updates for the OS/EE or 877 the security solution due to missing licenses. However the systems 878 may remain in use and become increasingly vulnerable as time goes on 879 and new attacks are discovered. 881 10.1.2. Mirai IoT bot 883 Editor's note: we are going to experiment a new model to represent 884 examples showing the limits of endpoint security solutions therefore 885 the first table is the old format and the new table prepares the new 886 format so that we ca develop 2 new I-Ds one for endpoint model in 887 terms of attack surface and the other one in terms of attack 888 landscape and potential attack orchestrations. In this I-D we will 889 glue all the dots and describe the defense orchestration, yet, based 890 on a normative approach to terminology used so that we don't need to 891 describe it here 893 +-------------+-----------------------------------------------------+ 894 | Description | A Mirai bot infecting various IoT devices through | 895 | | weak passwords over Telnet port TCP 23 and by using | 896 | | various vulnerabilities, for example the SonicWall | 897 | | GMS XML-RPC Remote Code Execution Vulnerability | 898 | | (CVE-2018-9866) on TCP port 21009. Once a device is | 899 | | compromised it will scan for further victims and | 900 | | then start a DoS attack. | 901 +-------------+-----------------------------------------------------+ 902 | Simplified | Compromised device scans network for multiple open | 903 | attack | ports, attempts infection through weak password and | 904 | process | exploits, downloads more payload, starts DoS | 905 | | attack. | 906 | | | 907 | UE | No security tool present on majority of IoT | 908 | | devices, hence no detection possible. If a | 909 | | rudimentary security solution with limited | 910 | | capabilities such as outgoing firewall is present | 911 | | on the IoT device e.g. router, then it might be | 912 | | able to detect the outbound DoS attack and slow it | 913 | | down. | 914 | | | 915 | References | [MIRAI1][MIRAI2] | 916 +-------------+-----------------------------------------------------+ 917 +---------------+---------------------------------------------------+ 918 | Name | Mirai | 919 +---------------+---------------------------------------------------+ 920 | Description | A device infection for participation into a | 921 | | botnet activity | 922 | | | 923 | Endpoint | IoT Devices | 924 | Targeted | | 925 | | | 926 | Attack | Telnet remote access; Weak default and existing | 927 | Surface | passwords; Code vulnerabilities in exposed | 928 | Categories | services | 929 | Involved | | 930 | | | 931 | Attack | Weak passwords over Telnet TCP port 23; SonicWall | 932 | Surface | GMS XML-RPC Remote Code Execution Vulnerability | 933 | Examples | (CVE-2018-9866) on TCP port 21009 | 934 | | | 935 | Attack | Deployment of a custom code or commands on the | 936 | Objective | device for participation in botnet activities | 937 | | | 938 | Attack | Botnet Deployment; DDoS | 939 | Category | | 940 | | | 941 | Attack | Exploit remote access weaknesses on the device to | 942 | Orchestration | deploy a bot on the device | 943 | | | 944 | Mitigation | If a rudimentary security solution with limited | 945 | | capabilities such as outgoing firewall is present | 946 | | on the IoT device e.g. router, then it might be | 947 | | able to detect the added bot or the outbound DoS | 948 | | attack and slow it down | 949 | | | 950 | Attack | Better password management; Uptodate patching | 951 | Surface | | 952 | Minimisation | | 953 | | | 954 | References | [MIRAI1][MIRAI2] | 955 +---------------+---------------------------------------------------+ 957 10.2. Endpoints may not see the malware on the endpoint 959 10.2.1. LoJax UEFI rootkit 960 +-------------+-----------------------------------------------------+ 961 | Description | A device compromised with the LoJax UEFI rootkit, | 962 | | which is active before the OS/EE is started, hence | 963 | | before the endpoint security is active. It can pass | 964 | | back a clean 'image' when the security solution | 965 | | tries to scan the UEFI. Infection can either happen | 966 | | offline with physical access or through a dropper | 967 | | malware from the OS/EE. | 968 +-------------+-----------------------------------------------------+ 969 | UE | A perfect endpoint security could potentially | 970 | | detect the installation process if it is done from | 971 | | the OS/EE and not with physical modification or in | 972 | | the factory. Once the device is compromised the | 973 | | endpoint security solution can neither detect nor | 974 | | remove the rootkit. The endpoint solution may | 975 | | detect any of the exhibited behaviour, for example | 976 | | if the rootkit drops another malware onto the OS/EE | 977 | | at a later stage. | 978 | | | 979 | Reference | [LOJAX] | 980 +-------------+-----------------------------------------------------+ 982 10.2.2. SGX Malware 984 +-------------+-----------------------------------------------------+ 985 | Description | Malware can hide in the Intel Software Guard | 986 | | eXtensions (SGX) enclave chip feature. This is a | 987 | | hardware-isolated section of the CPU's processing | 988 | | memory. Code running inside the SGX can use return- | 989 | | oriented programming (ROP) to perform malicious | 990 | | actions. | 991 +-------------+-----------------------------------------------------+ 992 | UE | Since the SGX feature is by design out of reach for | 993 | | the OS/EE, an endpoint security solution can | 994 | | neither detect nor remove any injected malware. A | 995 | | perfect endpoint security solution could | 996 | | potentially detect the installation process if it | 997 | | is done from the OS/EE and not with physical | 998 | | modification or in the factory. | 999 | | | 1000 | References | [SGX1] [SGX2] | 1001 +-------------+-----------------------------------------------------+ 1003 10.2.3. AMT Takeover 1004 +-------------+-----------------------------------------------------+ 1005 | Description | A targeted attack group can remotely execute code | 1006 | | on a system through the Intel AMT (Active | 1007 | | Management Technology) vulnerability | 1008 | | (CVE-2017-5689) over TCP ports 16992/16993. This | 1009 | | provides full access to the computer, including | 1010 | | remote keyboard and monitor access. The attacker | 1011 | | can install malware, modify the system or steal | 1012 | | information. | 1013 +-------------+-----------------------------------------------------+ 1014 | UE | The AMT is accessible even if the PC is turned off. | 1015 | | Therefore any endpoint security software installed | 1016 | | on the OS, would not be able to see this traffic | 1017 | | and therefore also not able to detect it. | 1018 | | | 1019 | References | [AMT1], [AMT2] | 1020 +-------------+-----------------------------------------------------+ 1022 10.2.4. AMT case study (anonymised) 1024 An enterprise has a data center containing very sensitive data. 1025 Their workstations use a certain Intel chipset which integrates the 1026 AMT feature for remote computer maintenance. AMT is an interface for 1027 hardware management of the workstations, including transmission of 1028 screen content and keyboard and mouse input for remote maintenance. 1029 Communication with the management workstation is implemented by AMT 1030 through the network interface card (NIC) on the motherboard. The 1031 network packets generated in this way are invisible both to the main 1032 processor and thus to the OS running on the workstation. In autumn 1033 of 2015, it became known that some AMT-enabled computers had a flaw 1034 that allowed AMT's remote maintenance component to be activated and 1035 configured by attackers. This also worked when the workstations were 1036 switched off. The leakage of data through this vulnerability is 1037 elusive and difficult to detect. The identified threat situation led 1038 the organization to a new requirement implementing a method that can 1039 reliably detect this and similar vulnerabilities. In particular, the 1040 detection of rootkits and manipulated firmware, and this includes 1041 also (UEFI) BIOS - has also been a focus of their attention. 1043 The method used as a solution, compares the desired data packets 1044 generated by a client operating system - the user, with the data 1045 packets received on the switch port. If more data has been received 1046 on the switch port than was been sent by the operating system - the 1047 user, there is a strong possibility that something bad is happening - 1048 like for example an infection via modified firmware or by rootkit. 1050 10.2.5. Users bypass the endpoint security 1052 +-------------+-----------------------------------------------------+ 1053 | Description | Endpoint security systems should not interfere with | 1054 | | the normal operation of the endpoint to the extent | 1055 | | that users become frustrated and want to disable | 1056 | | them or configure them to disable a significant | 1057 | | fraction of important security capabilities. | 1058 +-------------+-----------------------------------------------------+ 1059 | UE | Add-on endpoint security is now bypassed or | 1060 | | disabled by the user. Unless the endpoint is under | 1061 | | monitored management or can prevent a user from | 1062 | | modifying the configuration, then this is shutting | 1063 | | down a significant fraction of the security | 1064 | | capabilities. | 1065 | | | 1066 | References | [NINESIGNS] | 1067 +-------------+-----------------------------------------------------+ 1069 10.3. Endpoints may miss information leakage attacks 1071 Another aspect that endpoint security has issues in detecting are 1072 information disclosure or leakage attacks, especially on shared 1073 virtual/physical systems. 1075 10.3.1. Meltdown/Specter 1077 The Meltdown/Specter vulnerabilities and all its variants may allow 1078 reading of physical memory belonging to another virtual machine (VM) 1079 on the same physical system. This could reveal passwords, 1080 credentials, certificates etc. The trick is that an attacker can 1081 spin up his own VM on the same physical hardware. As this VM is 1082 controlled by the attacker, they will ensure that there is no 1083 endpoint security that detects the Meltdown exploit code when run. 1084 It is very difficult for the attacked VM to detect the memory read- 1085 outs. For know CPU vulnerabilities there are software patches 1086 available than can be applied. If it is an external service 1087 provider, it might not be in the power of the user to patch the 1088 physical system or to determine if this has been done by the 1089 provider. 1091 10.3.2. Network daemon exploits 1093 Other attack types, which leak memory data from a vulnerable web 1094 server, are quite difficult to detect for an endpoint security. For 1095 example the Heartbleed bug allows anyone on the Internet to read the 1096 memory of the systems protected by the vulnerable versions of the 1097 OpenSSL software. This could lead to credentials or keys being 1098 exposed. An endpoint solution needs to either patch the vulnerable 1099 application or monitor it for any signs of exploitation or data 1100 leakage and prevent the data from being exfiltrated. 1102 10.3.3. SQL injection attacks 1104 A SQL injection attack is an example of an attack that exploits the 1105 backend logic of an application. Typically this is a web application 1106 with access to a database. By encoding specific command characters 1107 into the query string, additional SQL commands can be triggered. A 1108 successful attack can lead to the content of the whole database being 1109 exposed to the attacker. There are other similar attacks that can be 1110 grouped together for the purpose of this task, such as command 1111 injection or cross site scripting (XSS). Although they are different 1112 attacks, they all at their core fail at input filtering and 1113 validation, leading to unwanted actions being performed. 1115 Applications that are vulnerable to SQL injections are very common 1116 and are not restricted to web applications. An endpoint solution 1117 needs to monitor all data entered into possible vulnerable 1118 applications. This should include data received from the network. A 1119 generic pattern matching for standard SQL injection attack strings 1120 can be applied to potentially block some of the attacks. In order to 1121 block all types of SQL injection attacks the endpoint solution should 1122 have some knowledge about the logic of the monitored application, 1123 which helps to determine how normal requests differ from attacks. 1124 Applications can be analysed at source code level for potential 1125 weaknesses, but dynamically patching is very difficult. See [SQL] 1127 10.3.4. Low and slow data exfiltration 1129 An endpoint security solution can detect low and slow data 1130 exfiltration, for example when interesting data sources are tracked 1131 and access to them is monitored. If the data source is not on the 1132 endpoint itself, e.g. a database in the network, then the received 1133 data needs to be tagged and its further use needs to be tracked. To 1134 make detection difficult, an attacker could decide to use an 1135 exfiltration process that sends only 10 bytes every Sunday to a 1136 legitimate cloud service. If that is not in the normal behavior 1137 pattern, then this anomaly could be detected by the endpoint. If the 1138 process that sends the data or the destination IP address have a bad 1139 reputation, then they could be stopped. Though it is very difficult 1140 to reliably block such an attack and most solutions have a specific 1141 threshold that needs to be exceeded before it is detected as an 1142 anomaly. 1144 10.4. Suboptimality and gray areas 1146 10.4.1. Stolen credentials 1148 Stolen credentials and misuse of system tools such as RDP, Telnet or 1149 SSH are a valid scenario during attacks. An attacker can use stolen 1150 credentials to remotely log into a system and access data or execute 1151 commands in this context like the legitimate user might do. An 1152 endpoint security solution can restrict access from specific IP 1153 addresses, but this is difficult in a dynamic environment and when an 1154 attacker might have already compromised a trusted device and misuse 1155 it as a stepping stone for lateral movement. The endpoint could 1156 perform additional checks of the source device, such as verifying 1157 installed applications and certain conditions. Again this will not 1158 work in all scenarios, e.g. a hijacked valid device during lateral 1159 movement. 1161 This means that the system will not be able to simply block the 1162 connection if the authentication with the stolen credentials 1163 succeeds. A multi factor authentication (MFA) could limit the use of 1164 stolen credentials, but depending on the system used and the 1165 determination of the attacker they might be able to bypass this 1166 hurdle as well e.g. cloning a SIM card to read text message codes. 1168 As a next step, a solution on the endpoint can monitor the behavior 1169 of the logged in user and determine if it represents expected normal 1170 behavior. Unfortunately, there is the chance for false positives 1171 that might block legitimate actions, hence the rules are usually not 1172 applied too tightly. The system can monitor for suspicious behavior, 1173 similar to malware detection, where every action is carefully 1174 analyzed and all activity is tracked. For example if the SSH user is 1175 adding all files to archives with passwords and then deletes the 1176 original files in the file explorer, then this could result in a 1177 ransomware case scenario. If only a few files are processed per 1178 hour, then this activity will be very difficult for the endpoint to 1179 distinguish from normal activity, in order to flag it as malicious. 1181 The problem of attackers blending in with normal activity is one of 1182 the biggest challenges with so called living off the land attack 1183 methods. The attacker chooses to keep their profile low by not 1184 installing any additional binary files on the system, but instead 1185 misuses legitimate system tools to carry out their malicious intent. 1186 This means that there is no malware file that could be identified and 1187 the detection relies solely on other methods such as behaviour based 1188 monitoring [LOTLSYMC]. 1190 If information is shared across multiple endpoints, then each one 1191 could learn from the others and see how many connections came in from 1192 that source, what files were involved and what behavior the clients 1193 exhibited. This crowd wisdom approach would allow blocking rules to 1194 be applied after the first incident across multiple endpoints. 1196 10.4.2. Zero Day Vulnerability 1198 +-------------+-----------------------------------------------------+ 1199 | Description | An attacker exploits a zero day vulnerability or | 1200 | | any recent vulnerability. | 1201 +-------------+-----------------------------------------------------+ 1202 | UE | In theory this scenario could be handled by the | 1203 | | endpoint security: a) Once the intrinsic security | 1204 | | system has been patched, exploitation of the | 1205 | | vulnerability can be prevented. b) The add-on | 1206 | | security with enhanced capabilities or updated | 1207 | | methods can detect and mitigate the vulnerability. | 1208 | | It does not necessarily require the official patch. | 1209 | | | 1210 | Challenge | In practice many systems remain vulnerable to a | 1211 | | vulnerability months or even years after a security | 1212 | | fix has been released. Moreover there is a big gap | 1213 | | between when a vulnerability is disclosed and when | 1214 | | a security fix is available. Also there is a big | 1215 | | gap between when a security fix is available and | 1216 | | when the security fix is actually applied. A recent | 1217 | | study over three years, examined the patching time | 1218 | | of 12 client-side and 112 server-side applications | 1219 | | in enterprise hosts and servers. It took over 6 | 1220 | | months on average to patch 90% of the population | 1221 | | across all vulnerabilities. [NDSSPATCH]. We note | 1222 | | too: "The patching of servers is overall much worse | 1223 | | than the patching of client applications. On | 1224 | | average a server application remains vulnerable for | 1225 | | 7.5 months." | 1226 | | | 1227 | References | [ZERODAY1][ZERODAY2] | 1228 +-------------+-----------------------------------------------------+ 1230 10.4.3. Port scan over the network 1232 An infected machine, let's say a Mirai bot on a router, is scanning a 1233 class B network for IP addresses with TCP port 80 open. The malware 1234 can slow it down to 1 IP address per 5 seconds (or any other 1235 threshold) and it can go in randomized order (like for example the 1236 nmap tool does) in order to make it difficult to find a sequential 1237 pattern. To increase detection difficulties, legitimate requests to 1238 existing web servers can be added in at random intervals. 1240 An endpoint solution might be able to detect this behaviour, 1241 depending on the threshold, but it will be difficult. At some point 1242 the pattern will be similar to browsing the web, so either the 1243 endpoint blocks the bot scanning and also the user from surfing, or 1244 it allows both. 1246 To make it even harder, the attacker can use a botnet that 1247 communicates over peer-to-peer (P2P) or a central command and control 1248 server (C&C) and then distribute the scan load over multiple hosts. 1249 This means each endpoint only scans a subset, let's say 100 IP 1250 addresses, but all 1,000 bots scan a total of 100,000 IP addresses. 1252 This attack is difficult to detect by a reasonable threshold on each 1253 endpoint individually. If the endpoints talk to each other and 1254 exchange information, then a collective decision can be made on the 1255 bigger picture of the bot traffic. 1257 Another option for the endpoint solution is to block the bot malware 1258 from operating on the computer, for example by detecting the 1259 installation, analyzing the behavior of the process or by preventing 1260 the binary from accessing the network. This includes blocking any 1261 form of communication for the process to its C&C server, regardless 1262 of if it is using a P2P network or misusing legitimate system tools 1263 or browsers to communicate with the Internet. Blocking indirect 1264 communication over system tools as part of living off the land 1265 tactics, can be very challenging. 1267 See [BOT] 1269 10.4.4. DDoS attacks 1271 For this example let us consider a botnet of 100,000 compromised 1272 computers and each one sends a burst of traffic to a remote target, 1273 for one second each, alternating in groups. This will generate some 1274 waves of pulse attack traffic. Similar comments can be made about 1275 overall pulsed DDoS attacks [PDDoS]. 1277 A solution on the endpoint can attempt to detect the outgoing 1278 traffic. If the DoS attack is volume based and the time span of each 1279 pulse is large enough or the repeating frequency for each bot is 1280 high, then detection with thresholds on the endpoint is feasible. It 1281 is different, if it is an application layer DoS attack, where the 1282 logic of the receiving application is targeted, for example with too 1283 many search queries in HTTP GET requests. This would flood the 1284 backend server with intensive search requests, which can result in 1285 the web site no longer being responsive. Such attacks can succeed 1286 with a low amount of requests being sent, especially if its 1287 distributed over a botnet. This makes it very difficult for a single 1288 endpoint to detect such an ongoing attack, without knowledge from 1289 other endpoints or the network. 1291 Another option for the endpoint solution is to block the bot malware 1292 from operating on the computer, for example by detection the 1293 installation, analyzing the behavior of the process or by preventing 1294 the binary from accessing the network. This includes blocking any 1295 form of communication for the process to its C&C server, regardless 1296 of if it is using a P2P network or misusing legitimate system tools 1297 or browsers to communicate with the Internet. Blocking indirect 1298 communication over system tools as part of living off the land 1299 tactics, can be very challenging. 1301 11. Learnings from production data 1303 From the above limited considerations we can now check what we see 1304 from real production data using 1306 o the method described in [MONEYBALL] 1308 o the anonymised production data of Symantec MSS production for the 1309 past 3 months 1311 The core idea is to consider, based on all the imperfections we 1312 started to list above including the 'grey areas', that cybersecurity 1313 analysts are often presented with suspicious machine activity that 1314 does not conclusively indicate a compromise, resulting in undetected 1315 incidents or costly investigations into the most appropriate remedial 1316 actions. 1318 As Managed Security Services Providers (MSSP's) are confronted with 1319 these data quality issues, but also possess a wealth of cross-product 1320 security data that enables innovative solutions, we decided to use 1321 the Symantec MSS service for the past 3 months. The Symantec MSS 1322 service monitors over 100 security products from a wide variety of 1323 security vendors for hundreds of enterprise class customers from all 1324 verticals. 1326 We selected the subset of customers using the service that deploy 1327 both network and endpoint security products to determine which types 1328 of security incidents were most likely to be detected by endpoint 1329 products vs. network products. In doing so, we were particularly 1330 interested in identifying which categories of incidents are detected 1331 by endpoint products and not network products, and vice versa. Thus, 1332 we examined prevalent categories of incidents for which the only 1333 actionable security alerts were predominantly produced by one type of 1334 security product and not the other. To do so, we extracted all 1335 security incidents detected by Symantec MSS on behalf of hundreds of 1336 customers that deploy both network and endpoint security products, 1337 over a three-month period from December 2018 through the end of 1338 February 2019. We acknowledge that some attacks might have been 1339 blocked by the first product and therefore have never been seen by 1340 the next security solution, which influences the final numbers. 1342 With this in mind, we could identify incidents based on: 1344 +------------+------------------------------------------------------+ 1345 | Severity | 4 - Emergency, 3 - Critical, 2 - Warning, 1 - | 1346 | | Informational | 1347 +------------+------------------------------------------------------+ 1348 | Incident | Malicious Code, Deception Activity, Improper Usage, | 1349 | Category | Investigation, etc. | 1350 | | | 1351 | Incident | Trojan Horse Infection, Suspicious DGA Activity, | 1352 | Type | Suspicious Traffic, Suspicious URL Activity, | 1353 | | Backdoor infection, etc. | 1354 | | | 1355 | # network | Amount of network only security incidents | 1356 | incidents | | 1357 | | | 1358 | # all | What is the total amount of incidents on all | 1359 | incidents | security solutions | 1360 | | | 1361 | Percentage | Percentage of network security only incidents | 1362 +------------+------------------------------------------------------+ 1364 We ended up with 1366 o Hundreds of thousands of security incidents 1368 o which we could categorize in 275 incident types by category and 1369 severity (triplets Severity-Category-Type) 1371 o out of which we searched how many incidents of each type were 1372 detected by a network security product and missed by deployed 1373 endpoint security products at least 75% of the time or vice versa 1375 11.1. Endpoint only incidents 1377 The categories of incidents that are detected primarily by endpoint 1378 security products are fairly intuitive. They consist primarily of 1379 detections of file-based threats and detection of malicious behaviors 1380 through monitoring of system and network behavior at the process 1381 level. The most prevalent of these behavioral detections include 1382 detections of suspicious URLs based on heuristics and blacklists of 1383 IP addresses or domain names. Since most of these alerts are not 1384 corroborated by network products, it seems probable that the 1385 blacklists associated with network products tend to be more focused 1386 on attacks while host-based intrusion prevention system alerts focus 1387 more on malware command and control traffic. Most other behavioral 1388 detections at the endpoint provide alerts based on system behavior 1389 that is deemed dangerous and symptomatic of malicious intent by a 1390 malicious or infected process. The highest severity incidents 1391 detected on endpoints are instances of post-compromise outbound 1392 network behavior that are symptomatic of command and control 1393 communications traffic, but these did not show up as being primarily 1394 detected by endpoint products as they are frequently corroborated by 1395 network-based alerts. 1397 11.2. Security incidents detected primarily by network security 1398 products 1400 Perhaps less intuitive are the results of examining categories of 1401 security incidents that are detected primarily by network security 1402 products and only rarely corroborated by endpoint security products. 1403 Below we provide details regarding incident categories for which a 1404 network security product produced a detection and for which there 1405 were no actionable endpoint alerts for at least 75% of the incidents 1406 in the category. 1408 In our study we found 32 incident type, category, and severity 1409 triplets of this type. The following categories critical incident 1410 types were reported by MSS customers, and we discuss each in turn in 1411 decreasing order of prevalence: 1413 11.2.1. Unauthorized external vulnerability scans 1415 Perhaps unsurprisingly, unauthorized external attempts to scan 1416 corporate resources for vulnerabilities and other purposes are 1417 detected in large volumes by a broad variety of network-focused 1418 security products. 79% of incidents of this type were detected by 1419 network security products with critical-severity alerts, these 1420 security incident detections are not accompanied by any actionable 1421 endpoint alerts, despite the fact that endpoint security products are 1422 deployed by these enterprises. This category of threats encompasses 1423 a broad variety of attacks, the most prevalent of which are the 1424 following: Horizontal scans, SQL injection attacks, password 1425 disclosure vulnerabilities, directory traversal attacks, and 1426 blacklist hits. Of these categories of detections, horizontal scans 1427 stand out as the category of detection that endpoint-security 1428 products are least likely to detect on their own. 1430 11.2.2. Unauthorized internal vulnerability scans 1432 Unauthorized internal vulnerability scans, though less frequent, are 1433 more alarming, as they are likely to represent possible post- 1434 compromise activity. We note that the Managed Security Service works 1435 with its customers to maintain lists of devices that are authorized 1436 to perform internal vulnerability scans, and their activity is 1437 reported separately at a lower levels of incident severity. 89% of 1438 detected unauthorized internal vulnerability scans are detected by 1439 network products without any corroborating actionable alerts from 1440 endpoint security products. As compared to unauthorized external 1441 scan incidents, internal hosts that perform vulnerability scans are 1442 far more active and the fraction of alerts that detect horizontal 1443 scans is higher, representing half of the total alerts generated. 1444 Alerts focused on Network-Behavior Anomaly Detection also appear for 1445 internal hosts. 1447 11.2.3. Malware downloads resulting in exposed endpoints 1449 This category of threats is generally detected by network security 1450 appliances. Despite these enterprises being purchasers of endpoint 1451 security products, 76% of the incidents detected by the network 1452 security products do not show a corresponding alert by an endpoint 1453 security product. A broad variety of network appliances contributed 1454 to the detection of a diverse collection of malware samples. 1456 11.2.4. Exploit kit infections 1458 This category of infections represents instances in which the 1459 customer's machines are exposed to exploit kits. These threats were 1460 detected by network appliances that extract suspicious URLs from 1461 network traffic taps and use a combination of sandbox technology and 1462 blacklists to identify websites that deploy a variety of exploit kits 1463 that were not being caught by endpoint security products. In this 1464 three month time period, the most prevalent categories of exploit 1465 kits detected involved redirections to the Magnitude exploit kit and 1466 exploit kits associated with phishing scams and attempts to expose 1467 users to fake Anti-Virus warnings and tools. A breakdown of the 1468 results is included below: 1470 +---------------------+-----------------------+ 1471 | Severity | 3 - Critical | 1472 +---------------------+-----------------------+ 1473 | Incident Category | Malicious Code | 1474 | | | 1475 | Incident Type | Exploit Kit Infection | 1476 | | | 1477 | # network incidents | 26 | 1478 | | | 1479 | # all incidents | 26 | 1480 | | | 1481 | Percentage | 100% | 1482 +---------------------+-----------------------+ 1484 The network security product that detected these incidents produced 1485 the following alerts: 1487 o Advanced Malware Payloads 1489 o Exploit.Kit.FakeAV 1491 o Exploit.Kit.Magnitude 1493 o Exploit.Kit.MagnitudeRedirect 1495 o Exploit.Kit.PhishScams 1497 o HTMLMagnitudeLandingPage 1499 11.2.5. Attacks against servers 1501 In addition to detecting the aforementioned critical security 1502 incident categories, network security devices frequently detect a 1503 broad variety of attacks against servers that usually lack 1504 corroboration at the endpoint. Most server attacks are not matched 1505 by endpoint protection alerts: 62% are unmatched for critical 1506 incidents, and 88% are unmatched as lower severity incidents. This 1507 category of incidents is the most prevalent category of incidents 1508 detected primarily by network products, but they are usually rated 1509 lower in severity than the aforementioned classes of alerts as they 1510 are very commonplace. Even when these alerts are corroborated by 1511 endpoint protection alerts, the endpoint alerts are often low in 1512 severity, as in the case of file-based threats that appear to have 1513 been blocked or successfully cleaned up by an Anti-Virus solution. 1514 The challenge in taking action against server attacks is that it can 1515 be difficult to assess which of these attacks were successful in 1516 causing actual damage, and for this reason, for the fraction of 1517 server attacks that demonstrate corroborating endpoint security 1518 alerts, even if of low severity, should be examined. It is 1519 interesting to note the cooperative role played by both network and 1520 endpoint security devices in these instances. 1522 12. Regulatory Considerations 1524 This section will briefly look at the regulatory landscape and 1525 develop a specific view on the impact on endpoints with the goal to 1526 see what we might be able to learn. 1528 Legal requirements, compliance, regulatory frameworks and mandatory 1529 reporting are no longer separate from any security evaluation, 1530 process or requirement within an organisation, enterprise system or 1531 intranet. It is essential to look at the technical and regulatory 1532 approaches together. This section will look at two examples of legal 1533 requirements and guidance: 1535 (1) IoT security (2) Network infrastructure 1537 This section is by no means complete, but it does a discussion on 1538 this aspect of endpoint and ecosystem regulation. 1540 12.1. IoT Security 1542 IoT security regulation is emerging in the form of voluntary 1543 frameworks and self-assessments that relate to endpoint security 1544 issues.. These frameworks focus first on the end point, or mobile 1545 device, in the IoT environment and then on the holistic ecosystem 1546 itself. 1548 In 2017 the National Institute of Standards and Technology released 1549 its draft IoT Cybersecurity Framework based on consultations and 1550 interviews with all stakeholders over several years previously 1551 [NISTIOTP]. Some of the themes which emerged was the need for IoT 1552 governance, assessment frameworks, review of all aspects of the IoT 1553 ecosystem and a process for coordinated vulnerability disclosure 1554 inside an organisation. As evidenced by the 2018 Endpoint Protection 1555 and Response Survey by SANS, only 47% of organisations know that 1556 their endpoints have been breached and a further 20% are unsure 1557 [EPRSANS]. So a systemic approach from NIST was welcomed and the 1558 NIST framework became the gold standard for national IoT security 1559 frameworks. 1561 Other IoT security frameworks include the Singapore IoT Cyber 1562 Security Guide from January 2019 and the UK's Secure by Design or The 1563 Government's Code of Practice for Consumer Internet of Things (IoT) 1564 Security for manufacturers, with guidance for consumers on smart 1565 devices at home [IMDAIOTG], [SBDGOVUK]. Once again both look at 1566 securing the IoT device or endpoint, but also security for the entire 1567 value chain of the IoT system. The Singapore framework makes the 1568 point about the entire system clear, "Similar to any system, an IoT 1569 system is as secure as its weakest link. It is thus important to 1570 ensure that proper security considerations and measures are put in 1571 place for both the implementation and operational stages of the 1572 deployment of any IoT system." [IMDAIOTG] Finally, the IoT Security 1573 Foundation, the GSMA and the Internet Society have all released their 1574 own frameworks for IoT security. All have similar characteristics 1575 which focus on the entire value chain and ecosystem, but also on 1576 vulnerability disclosure and checklist assessments. What makes each 1577 of these approaches slightly different is the differing perspectives 1578 of the organization advocating it. The GSMA is the mobile trade 1579 association and so it focuses on mobile devices while the Internet 1580 Society focuses on the Internet ecosystem and a multistakeholder 1581 approach. Systematically underpinning all the frameworks is the 1582 holistic approach with voluntary best practices and implementation 1583 based on the needs of the user or organisation adopting the framework 1584 [IOTSFCF], [GSMAIOT], [ISTRUST]. 1586 12.2. Network infrastructure 1588 In Europe, the Network and Information Security Directive, which was 1589 passed in July 2016, require implementation by each European member 1590 state with a threefold aim. First, to put into place a national 1591 strategy for network and infrastructure security including best 1592 practices, guidelines, training and stakeholder consultations. 1593 Second, to coordinate national CSIRTs with CERT-EU and third to 1594 provide incident control and response systems for critical 1595 infrastructure and digital services [EURLEX]. This Directive 1596 demonstrates the importance give across the EU to network resilience 1597 and incident reporting. While securing the endpoint is acknowledged, 1598 the focus is on ensuring the security of European interoperable 1599 networks. In short, the importance of the security of the network 1600 including incident response shows that it isn't only the endpoints 1601 that should be the focus of the regulation and legal frameworks. 1603 12.3. Auditing and Assessment 1605 This section will talk about other risk assessment and auditing 1606 regulatory requirements beyond the NIS directive. 1608 One example of risk assessment as a regulatory requirement is the New 1609 York State law 23 NYCRR 500 of the Regulations of the Superintendent 1610 of Financial Services (Cybersecurity Requirements for Financial 1611 Services Companies). Among the requirements, audit, risk assessment 1612 and risk reporting are included like, 1613 (2) include audit trails designed to detect and respond to 1614 Cybersecurity Events that have a reasonable likelihood of materially 1615 harming any material part of the normal operations of the Covered 1616 Entity. [NYCYBER] 1618 12.4. Privacy Considerations 1620 We may consider a specific focus on privacy in the future. 1622 13. Human Rights Considerations 1624 This section may develop a specific view of requirements, limits and 1625 constraints coming from Human Rights perspective on endpoint 1626 security. 1628 14. Security Considerations 1630 This document is about Security Considerations 1632 15. IANA Considerations 1634 This document has no actions for IANA 1636 16. Informative References 1638 [ADAPTURE] 1639 Cullen, T., "Limits of endpoint only", July 2017, 1640 . 1643 [AMT1] Khandelwal, S., "Explained - How Intel AMT Vulnerability 1644 Allows to Hack Computers Remotely", May 2017, 1645 . 1648 [AMT2] Symantec, ., "Web Attack Intel AMT Privilege Escalation 1649 CVE-2017-5689", 2017, 1650 . 1653 [ATTACK] "MITRE ATT&CK", n.d., . 1655 [BOT] Marinho, R., "Exploring a P2P transient botnet - From 1656 Discovery to Enumeration", July 2017, 1657 . 1660 [CANDID1] Wueest, C., "How my TV got infected with ransomware and 1661 what you can learn about it", November 2015, 1662 . 1665 [CANDID2] Dickson, B., "Millions of smart TVs are vulnerable to 1666 hackers", February 2014, 1667 . 1669 [CAPEC] "MITRE CAPEC", n.d., 1670 . 1672 [ENISA] ENISA, ., "Baseline Security Recommendations for IoT in 1673 the context of Critical Information Infrastructures", 1674 November 2017, . 1677 [EPPEDR] Redscan, ., "EPP and EDR - What's the difference?", June 1678 2018, . 1681 [EPPGUIDE] 1682 "IT Pro's Guide to Endpoint Protection", n.d., 1683 . 1686 [EPPSECURITY] 1687 Hunt, J., "Advantages and Disadvantages of Three Top 1688 Endpoint Security Vendors", n.d., 1689 . 1692 [EPRSANS] Neely, L., "Endpoint Protection and Response A SANS 1693 Survey", June 2018, . 1696 [EPTAXONOMY] 1697 MacFadden, M., "Endpoint Taxonomy for CLESS", July 2019, 1698 . 1701 [ERICSSON] 1702 Ericsson, ., "Internet of Things forecast", n.d., 1703 . 1706 [EURLEX] EUP, ., "Directive (EU) 2016/1148", July 2016, 1707 . 1710 [FLAMER] Symantec, ., "W32.Flamer Microsoft Windows Update Man-in- 1711 the-Middle", June 2012, 1712 . 1715 [GARTNERIOT] 1716 Van der Meulen, R., "Gartner Says 8.4 Billion Connected 1717 Things Will be in Use in 2017, Up 31 percent from 2016", 1718 February 2017, . 1722 [GARTNERREPORT] 1723 Crotty, J., "New Gartner Report Redefines Endpoints 1724 Protection for 2018", January 2018, 1725 . 1728 [GSMAIOT] GSMA, ., "GSMA IoT Security Guidelines and Assessment", 1729 n.d., . 1732 [HSTODAY] Hstoday, ., "Layered Approach Critical to Effective 1733 Endpoint Protection", October 2016, 1734 . 1738 [I-D.draft-mcfadden-smart-endpoint-taxonomy-for-cless-00] 1739 McFadden, M., "Endpoint Taxonomy for CLESS", draft- 1740 mcfadden-smart-endpoint-taxonomy-for-cless-00 (work in 1741 progress), July 2019. 1743 [IMDAIOTG] 1744 IMDA, ., "IMDA IoT Cyber Security Guide", January 2019, 1745 . 1750 [IOTPATCHING] 1751 Rogers, D., "Handling vulnerabilities as an IoT vendor", 1752 December 2018, . 1756 [IOTSFCF] IoTSF, ., "IoT Security Compliance Framework", December 1757 2018, . 1760 [ISTRUST] ISOC, ., "Internet of Things (IoT) Trust Framework v2.5", 1761 May 2018, 1762 . 1765 [LOJAX] ESET, ., "LoJax First UEFI rootkit found in the wild, 1766 courtesy of the Sednit group", September 2018, 1767 . 1770 [LOTLSYMC] 1771 Wueest, C., "Living off the land and fileless attack 1772 techniques", July 2017, 1773 . 1777 [MIRAI1] Symantec, ., "Mirai, what you need to know about the 1778 botnet behind recent major DDoS attacks", October 2016, 1779 . 1782 [MIRAI2] Krebsonsecurity, ., "19 Mirai Botnet Authors Avoid Jail 1783 Time", September 2018, 1784 . 1786 [MONEYBALL] 1787 Roundy, K., "Predicting Cyber Threats with Virtual 1788 Security Products. ACSAC", 2017, 1789 . 1792 [NDSSPATCH] 1793 Caballero, J., "Mind Your Own Business A Longitudinal 1794 Study of Threats and Vulnerabilities in Enterprises", 1795 February 2019, . 1799 [NETTODAY] 1800 Dix, J., "Layered Security Defenses What layer is most 1801 critical network or endpoint", July 2011, 1802 . 1806 [NINESIGNS] 1807 Smith, K., "9 signs your endpoint security isn't working", 1808 May 2017, . 1811 [NISTIOTP] 1812 NIST, ., "NIST Cybersecurity for IoT Program", November 1813 2016, . 1816 [NYCYBER] NYCRR, ., "See 3 NYCRR 500 of the Regulations of the 1817 Superintendent of Financial Services (Cybersecurity 1818 Requirements for Financial Services Companies)", n.d., 1819 . 1822 [OWASP] OWASP, ., "Defense in depth definition", August 2015, 1823 . 1825 [PDDoS] Seals, T., "Pulse-Wave DDoS Attacks Mark a New Tactics in 1826 Q2", October 2017, . 1829 [SBDGOVUK] 1830 UK, GOV., "Secure by Design", February 2019, 1831 . 1834 [SGX1] Claburn, T., "Intel SGX safe room easily trashed by white- 1835 hat hacking marauders Enclave malware demoed", February 1836 2019, . 1839 [SGX2] Cimpanu, C., "Researchers hide malware in Intel SGX 1840 enclaves", February 2019, . 1843 [SQL] Cobb, M., "SQL injection detection tools and prevention 1844 strategies", November 2009, 1845 . 1848 [STATISTA1] 1849 Statista, ., "Internet of Things (IoT) connected devices 1850 installed base worldwide from 2015 to 2025 (in billions)", 1851 n.d., . 1854 [STATISTA2] 1855 Statista, ., "Size of Internet of Things market worldwide 1856 in 2014 and 2020 by industry (in billion U.S dollars)", 1857 n.d., . 1861 [TEEP] Cam-Winget, N., "Trust Execution Environment Protocol", 1862 March 2018, . 1864 [USCERT] Michael, C., "Principles of defense-in-depth", September 1865 2005, . 1869 [ZERODAY1] 1870 McHugh, J., "Windows of Vulnerability A Case Study 1871 Analysis", 2000, . 1874 [ZERODAY2] 1875 Plattner, B., "Large-Scale Vulnerability Analysis", 1876 September 2006, . 1879 Appendix A. Contributors 1881 o Arnaud Taddei 1882 Broadcom 1883 arnaud.taddei@broadcom.com 1885 o Bret Jordan 1886 Broadcom 1887 bret.jordan@broadcom.com 1889 o Candid Wueest 1890 Acronis 1891 candid@wueest.ch 1893 o Chris Larsen 1894 Broadcom 1895 chris.larsen@broadcom.com 1897 o Andre Engel 1898 Broadcom 1899 andre,engel@broadcom.com 1901 o Kevin Roundy 1902 Norton Lifelock 1903 kevin.roundy@nortonlifelock.com 1905 o Yuqiong Sun 1906 Norton Lifelock 1907 Yuqiong.Sun@nortonlifelock.com 1909 o David Wells 1910 Symantec 1911 David_Wells@symantec.com 1913 o Dominique Lazanski 1914 Last Press Label 1915 dml@lastpresslabel.com 1917 Authors' Addresses 1919 Arnaud Taddei 1920 Broadcom 1921 1320 Ridder Park Dr 1922 San Jose CA 1923 USA 1925 Email: arnaud.taddei@broadcom.com 1927 Candid Wueest 1928 Acronis 1929 Rheinweg 9 1930 Schaffhausen 8200 1931 Switzerland 1933 Email: candid@wueest.ch 1935 Kevin A. Roundy 1936 Norton Lifelock 1937 60 E Rio Salado Pkwy STE 1000 1938 Tempe AZ 85281 1939 USA 1941 Email: Kevin.Roundy@nortonlifelock.com 1942 Dominique Lazanski 1943 Last Press Label 1944 Flat 1, 109A Columbia Road 1945 London E2 7RL 1946 UK 1948 Email: dml@lastpresslabel.com