Published July 31, 2026 | Version 3

Hardware-rooted attestation for AI-agent evidence: composing IETF RATS with action evidence packages

Authors/Creators

  • 1. Tyche Institute, Tallinn, Estonia

Description

Version 3 (31 July 2026) is a correction of version 2 (17 July 2026, archived under the same concept DOI). No claim, result, experiment, figure or measurement changed. Three corrections were applied:

  • Reference [4] corrected. It described a companion manuscript as submitted to IEEE Transactions on Technology and Society with the manuscript number pending. That manuscript was rejected upon initial review on 18 July 2026 — one day after this version 2 was deposited — so a statement that was accurate at deposit had since gone stale. The reference now cites the corresponding public data deposit, doi:10.5281/zenodo.20488643, which is stable and carries the underlying materials.
  • §7 count corrected. The text reported “three items deliberately left open”; with the manuscript number retired, two remain — the exact Veraison service and interface names, and the §5 result-schema-to-vocabulary mapping. Both are still marked in the text, as in earlier versions.
  • Declarations section added, stating generative-AI assistance, funding and competing interests. This brings the deposit into line with the version of this work submitted to a peer-reviewed venue, which already carried such a statement.

The notice that this work has been submitted to the IEEE for possible publication, present in version 2, is unchanged.

An action evidence package (AEP) is a signed, append-only record of what an AI agent did, who or what authorised the action, and what the outcome was. It is a software-layer artefact: it tells a verifier the story of an action as the agent's own runtime reports it. This note argues that software attestation of this kind is necessary but not sufficient. When a verifier's question shifts from 'what does the agent claim it did?' to 'did the specific model version the operator claims to have deployed actually produce this output, on unmodified hardware?', the AEP alone cannot answer. The missing element is a hardware root of trust. The IETF Remote Attestation Procedures (RATS) architecture (RFC 9334) and Veraison, an open-source RATS Verifier implementation (Confidential Computing Consortium / Linux Foundation), supply exactly this. We propose a composite attestation -- hardware Evidence appraised under RATS, bound to a software AEP -- and map a small verifier vocabulary (Authorised / Unauthorised / Indeterminate / Attested / Contested / Expired) onto RATS appraisal outcomes. A feasibility experiment using swtpm 0.7.3 demonstrates the binding and three-way platform verdict end to end: Attested for a good, fresh quote; Contested when the model measurement is swapped; Expired when a stale quote is replayed; and rejection of a forged AEP outcome bound to a valid quote.

Note. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.

Version 2 (2026-07-17): removed leftover editorial placeholders ([VERIFY: ...] markers) from the reference list and the stale target-venue line; body text unchanged from version 1.

Notes

This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Experiment uses swtpm 0.7.3. Companion to the author's evidence-instrumentation work (IEEE Transactions on Technology and Society, under review). Three items remain open by design and are marked [VERIFY] in the text: the companion manuscript number, exact Veraison service/interface names, and the section-5 result-schema to vocabulary mapping.

Files

hardware-rooted-attestation-aep-rats-sokolov-2026.pdf

Files (235.5 kB)

Name Size
md5:52075a505a723756c6d92d74998326a2
235.5 kB Preview Download

Additional details

Related works

Is identical to
arXiv:2608.00801 (arXiv)