Internet-Draft Chihna Validation Receipts November 2026
Murthy Expires 7 May 2027 [Page]
Workgroup:
Network Working Group
Published:
Intended Status:
Informational
Expires:
Author:
V. Murthy
Venture Vertex LLC

Chihna Validation Receipts: Portable Evidence of AI Agent Adversarial Testing

Abstract

Agent receipts are standardizing around what agents DID: execution receipts record tool calls, signed receipts record actions taken, and compliance receipts bind those actions to regulatory obligations. What none of them carry is what agents WITHSTOOD: the adversarial validation evidence that an agent, model, or deployment was tested against attack and what honestly happened.

This document defines the validation receipt: a portable, digest-bound summary of an adversarial validation campaign, joined to runtime behavior evidence. It composes with execution receipts rather than competing with them: the runtime half is theirs; the withstood half is this document. The design principle is radical checkability: no keys, no network, no trusted third party. SHA-256 and the rules in this document. Anyone can verify a validation receipt with a JSON parser and a hash function.

The profile's central rule is the honest-statistics rule: verdict counts MUST include inconclusive results, and counts that do not sum are non-conformant. A campaign judged by a deterministic analyzer routinely produces inconclusive verdicts; hiding them is the exact failure mode this profile exists to make impossible.

Status of This Memo

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 5 May 2027.

▲

Table of Contents

1. Introduction

Operators deploying AI agents increasingly require evidence that the deployed system was tested against adversarial input: red-team campaigns, jailbreak suites, refusal-marker analysis. Today that evidence is reported in prose, slide decks, and marketing claims. A relying party - an insurer, an auditor, a regulator, a customer - cannot check any of it.

Receipts work in progress elsewhere in the IETF records actions taken: [AER1] defines a portable execution receipt for tool calls, [XAIP] defines signed execution receipts, [ASQAV] defines a compliance profile of signed action receipts, and the SCITT architecture ([SCITT-ARCH]) provides a general claim transparency model. These answer "what did the agent do?". None of them answer "what was the agent tested against, and what honestly happened?".

This document defines the validation receipt, the portable summary of an adversarial validation campaign. It is a single JSON object that names its subject, states verdict counts under an honest-statistics rule, binds by SHA-256 digest to the underlying evidence artifacts, and joins to runtime behavior evidence via a hash-chained ledger head. It is deliberately implementable with no keys and no network: the current threat model is tamper detection and honest self-description, not impersonation resistance (see Section 8).

The name "chihna" (Sanskrit: mark, symbol) denotes the complete proof object of which the validation receipt is the portable summary: campaign artifacts, behavior ledger, and attestations. This document specifies only the receipt and its verification; the artifact formats it summarizes are described in terms of the bindings the receipt requires.

2. Requirements Language

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

3. Terminology

Campaign
A set of adversarial cases executed against a target under an explicit authorization basis, producing a case log and an evidence pack.
Evidence pack
The structured summary of a campaign: verdict counts, per-case verdicts, learning-loop flags.
Case log
The per-case record of prompts sent and responses received.
Behavior ledger
An append-only, hash-chained log of runtime actions (the execution-receipts half).
Attestation
A digest-bound statement binding an evaluation to its evidence pack.
Chihna
The complete proof object - campaign artifacts, ledger, and attestations - of which the validation receipt is the portable summary.
Emitter
An implementation that produces validation receipts. A conformant emitter can be written from this document alone.
Verifier
An implementation that checks validation receipts against the rules in this document and, when the underlying artifacts are available, against the artifacts themselves.

4. Receipt Data Model

A validation receipt is a single JSON object [RFC8259] conforming to the following structure. Field order is not significant. An example receipt:

{
  "profile": "chihna-validation-receipt/0.1",
  "subject": {
    "target": "openrouter-model",
    "model_id": "anthropic/claude-opus-5.5",
    "authorization_basis": "default: standard API terms of service"
  },
  "campaign": {
    "campaign_id": "frontier_passport-20261103-045718",
    "cases_total": 38,
    "cases_passed": 13,
    "cases_failed": 0,
    "cases_inconclusive": 25,
    "technique_classes": ["direct", "multi-turn", "encoding"],
    "evidence_pack_sha256": "<sha256 of the pack bytes>",
    "case_log_sha256": "<sha256 hex of the case log bytes>"
  },
  "runtime": {
    "ledger_agent": "chihna-phase1-opus",
    "ledger_entries": 39,
    "ledger_head_sha256": "<entry_hash of the ledger's last entry>"
  },
  "attestation": {
    "evaluation_sha256": "<sha256 the attestation binds to>",
    "continuity": true
  },
  "verification": {
    "verified": true,
    "issues": []
  }
}

The top-level members are:

profile
The profile identifier and version, a string. See Section 5 rule 1.
subject
What was tested: the target class, the model identifier, and the authorization basis under which testing was conducted.
campaign
The campaign identity, verdict counts, technique classes exercised, and the SHA-256 digests binding the receipt to the evidence pack and case log.
runtime
Optional. The behavior-ledger half: the ledger agent, entry count, and chain head. A receipt MAY omit this block entirely (campaign-only receipt).
attestation
Optional. The digests of attestations binding the evaluation to its evidence, and whether a continuity attestation is present.
verification
The issuer's own verification result at issuance. A claim, not a proof; see Section 5 rule 6.

5. Normative Rules

The following rules define conformance of a validation receipt. A receipt that violates any MUST or MUST NOT in this section is non-conformant, and a conformant verifier MUST reject it.

  1. profile MUST be exactly "chihna-validation-receipt/0.1" for this version. Verifiers MUST reject unknown profile versions.

  2. subject.model_id MUST identify the model under test. A receipt for a mock or free-tier target MUST carry that identity honestly; the receipt MUST NOT imply frontier-class testing that did not occur.

  3. The honest-statistics rule. The verdict counts MUST include cases_inconclusive. A receipt whose counts do not sum - passed + failed + inconclusive + not_sent = total - is non-conformant. A receipt that omits the inconclusive count is non-conformant. A campaign judged by a deterministic analyzer routinely produces inconclusive verdicts; hiding them is the exact failure mode this profile exists to make impossible.

  4. campaign.evidence_pack_sha256 MUST be the SHA-256 [FIPS180-4] of the evidence pack file bytes. A verifier with access to the pack MUST re-derive and compare; mismatch is a verification failure.

  5. runtime.ledger_head_sha256 MUST be the entry hash of the ledger's final entry. A receipt MAY omit the runtime block entirely; it MUST NOT present a partial or unchained ledger as continuity.

  6. verification records the issuer's own verification result at issuance. It is a claim, not a proof: a verifier MUST re-verify from the evidence rather than trust this block.

  7. A receipt is a SUMMARY. It does not contain the evidence pack, the case log, or the ledger. It names where they live (campaign_id) and binds to them by digest. Full verification requires the underlying artifacts.

6. Verification Algorithm

A conformant verifier of a validation receipt:

  1. Checks profile is a known version.

  2. Checks the count-sum rule (Section 5 rule 3).

  3. If the evidence pack is available: re-derives its SHA-256 and compares to campaign.evidence_pack_sha256; re-derives verdict counts from the pack and compares to the receipt's counts.

  4. If the ledger is available: walks the hash chain and checks the head against runtime.ledger_head_sha256.

  5. Reports each failure separately with its specific reason.

Digests bind but do not authenticate: a receipt from anyone can claim anything about digests they computed. This is deliberate and is treated in Section 8.

7. Relationship to Existing Receipts Work

This profile composes with, rather than competes against, the receipts work in progress in the IETF:

Execution receipts (AER-1) [AER1]
Complementary. AER-1 records actions taken; this profile records attacks withstood. A chihna can carry an AER-1 receipt as its runtime half.
Signed execution receipts (XAIP) [XAIP]
Complementary. XAIP signs the runtime half; this profile summarizes the validation half and binds both by ledger head.
Compliance receipts (ASQAV) [ASQAV]
Complementary. The campaign evidence maps to compliance packing (EU AI Act, NIST AI RMF, ISO/IEC 42001); this profile does not redefine that mapping.
SCITT [SCITT-ARCH]
The ledger chain and attestation digests align with SCITT's claim model; this profile is deliberately implementable without a transparency service, and compatible with one.

8. Security Considerations

Digests bind but do not authenticate: a receipt from anyone can claim anything about digests they computed. Keyed signing (for example, SCITT-style) is a future profile version, deliberately deferred until relying parties demand it. The current threat model is tamper detection and honest self-description, not impersonation resistance.

The receipt contains no secrets by construction (digests and counts only); emitters MUST NOT embed credentials, payloads, or API keys.

Verifiers MUST treat the verification block as an issuer's claim (Section 5 rule 6) and re-verify from the underlying artifacts when those artifacts are available.

9. IANA Considerations

This document has no IANA actions. The profile identifier chihna-validation-receipt/0.1 is a versioned string local to this specification; a future version MAY register a media type for the receipt format.

10. Normative References

[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/rfc/rfc2119>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/rfc/rfc8174>.
[RFC8259]
Bray, T., "The JavaScript Object Notation (JSON) Data Interchange Format", STD 90, RFC 8259, DOI 10.17487/RFC8259, , <https://www.rfc-editor.org/rfc/rfc8259>.
[FIPS180-4]
National Institute of Standards and Technology, "Secure Hash Standard (SHS)", NIST FIPS Publication 180-4, .

11. Informative References

[AER1]
Zambo, N., "AER-1: A Portable Execution Receipt for AI Agent Tool Calls", Work in Progress, Internet-Draft, draft-zambo-aer1-26, , <https://datatracker.ietf.org/doc/html/draft-zambo-aer1-26>. Work in progress.
[XAIP]
Kumakichi, K., "Signed Execution Receipts for AI Agent Tool Calls (XAIP Receipts)", Work in Progress, Internet-Draft, draft-xkumakichi-xaip-receipts-03, , <https://datatracker.ietf.org/doc/html/draft-xkumakichi-xaip-receipts-03>. Work in progress.
[ASQAV]
Marques, R., "Compliance Profile of Signed Action Receipts for AI Agents", Work in Progress, Internet-Draft, draft-marques-asqav-compliance-receipts-09, , <https://datatracker.ietf.org/doc/html/draft-marques-asqav-compliance-receipts-09>. Work in progress.
[SCITT-ARCH]
Internet Engineering Task Force, "An Architecture for Trustworthy and Transparent Digital Supply Chains", Work in Progress, Internet-Draft, draft-ietf-scitt-architecture-22, , <https://datatracker.ietf.org/doc/html/draft-ietf-scitt-architecture-22>. Work in progress. SCITT Working Group.

Author's Address

Vyasa Murthy
Venture Vertex LLC