| Internet-Draft | Agent Non-Collapse Requirements | October 2026 |
| Watts | Expires 11 April 2027 | [Page] |
This document defines architectural non-collapse requirements for agent systems that consume evidence, provenance, model outputs, or other machine-readable claims before consequential actions. It requires implementations to keep integrity, provenance, evidence qualification, authorization, and execution as distinct states; to represent unresolved required evidence explicitly; and to preserve issued evaluation records append-only. The document defines no new transport protocol and no IANA registrations.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 11 April 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Agent systems increasingly combine identity, delegation, evidence, memory, policy, and execution. A recurring failure mode is semantic collapse: a property established in one layer is treated as though it were established in another. Examples include treating a signed object as true, a supported premise as authorized action, or an authorization as proof of successful execution.¶
An implementation conforming to this architecture MUST distinguish at least the following states: artifact integrity, provenance, evidence qualification, authorization decision, and execution outcome.¶
Integrity MUST NOT be interpreted as proposition validity. Provenance MUST NOT be interpreted as truth. Evidence qualification MUST NOT be interpreted as authorization. Authorization MUST NOT be interpreted as execution.¶
Where a policy requires qualified evidence, missing, stale, unresolved, or type-incompatible required evidence MUST NOT be silently treated as satisfied.¶
Evidence evaluation SHOULD preserve PASS, FAIL, and INDETERMINATE as distinct states. FAIL indicates a required condition is contradicted or violated. INDETERMINATE indicates that no required condition has established FAIL but one or more required inputs or bridges are missing or unresolved. A policy requiring PASS MUST NOT accept INDETERMINATE.¶
Issued evaluation records SHOULD be immutable. New evidence SHOULD create a successor record that can identify its predecessor. A successor MUST NOT be interpreted as erasing the historical state under which an earlier decision was made.¶
Crossing from observation to evidence, evidence to warranted premise, premise to authorization, and authorization to execution requires explicit interfaces or witnesses. Implementations SHOULD make these promotion rules inspectable and testable.¶
An Evidence Qualification Receipt can implement the evidence-qualification layer described here. A PASS receipt remains an authorization input only; local policy retains authority over action.¶
Semantic collapse can enable authority laundering, replay of stale evidence, profile substitution, confused-deputy execution, and false completion records. Implementations should bind evidence states to exact propositions, profiles, contexts, and actions when consequence warrants it.¶
Evidence graphs can expose sensitive information. Systems should minimize plaintext evidence in authorization paths and prefer scoped digests or privacy-preserving references when sufficient.¶
This document has no IANA actions.¶