Written evidence submitted by AffixIO House of Lords International Relations and Defence Committee Second phase inquiry into Multilateral AI Governance Submission date: 20 November 2026 Deadline: 26 November 2026 About the submitter 1. AffixIO builds software that creates a checkable proof that a specific presence and action occurred: that X was there when X did this. Verification can be performed offline. It does not require an internet connection, an AffixIO account, or a call to AffixIO during the check. AffixIO is not seeking funding. AffixIO does not claim any Ministry of Defence contract, pilot, or endorsement. No classified work is described here. 2. This submission answers selected questions from the Committee's call for evidence where AffixIO has direct operational experience: Questions 8, 9, 10, 11, 14, 17 and 19. AffixIO does not offer evidence on Questions 1 to 7, 12, 13, 15, 16 or 18, which turn on geopolitical assessment beyond this submitter's remit. Question 8: Areas of AI best suited to multilateral governance 3. Attestation and verification are suited to multilateral treatment. Models and applications will continue to diverge across jurisdictions. Shared rules for how a machine-issued claim is verified, who may issue it, and who may dispute it are more tractable than shared rules for every model or every use case. 4. Arms-control style listing of model capabilities is hard without shared measurement and shared evidence formats. A procedure for post-quantum attested proofs of presence and action sits closer to standards and verification practice than to model regulation. Environmental and economic impact reporting may also suit multilateral work, but AffixIO's evidence here is limited to attestation and offline verification. 5. The practical gap is simple. To prove that a handover or similar event occurred, organisations today commonly either disclose a full operational log or require the verifier to query a central database. A central database can be hacked, leaked, or queried beyond the question that needed answering. A checkable proof that answers only the event question reduces that disclosure. Question 9: Factors that will determine cooperation over the next five years 6. The most important near-term factor is whether states agree a verification and dispute procedure for machine-issued attested claims, including post-quantum signatures. Without that procedure, acceptance or rejection of such a proof remains ad hoc even when the cryptography works. 7. Related factors are: (a) whether verification can be performed offline without dependence on a foreign hosted database at the moment of checking; (b) whether issuer authority is named in advance; and (c) whether reuse, alteration and forgery fail closed under independent test. Cooperation that ignores those three points will produce statements without an enforcement path. Question 10: Is the UK well placed to shape multilateral AI governance? 8. The United Kingdom is well placed on verification procedure because it already maintains analogous practices for passports and other official evidence: issuance authority, checking practice, and dispute routes. AffixIO's assessment is that the UK has not yet published an equivalent public procedure for a post-quantum attested proof of presence and action. 9. The UK's position between major technology and regulatory blocs is an advantage only if it can table a concrete verification method that others can run without joining a proprietary platform. Influence follows from a reusable procedure, not from hosting another forum. Question 11: Capabilities the UK needs to be an effective shaper 10. Required capabilities include: (a) technical ability to verify post-quantum attested artefacts offline; (b) diplomatic ability to propose issuer and dispute rules in standards and defence-adjacent fora; and (c) industrial ability to run non-classified pilots that produce artefacts others can replay. 11. Building those capabilities does not require AffixIO funding. It requires a named lead department for verification procedure, a non-classified pilot path, and publication of how a UK verifier should treat reuse, alteration and forgery. Current UK public and private investment levels in AI more broadly are [TO CONFIRM - DO NOT GUESS]. Question 14: Domestic regulatory alterations for international influence 12. The UK does not need a new AffixIO-specific statute for influence on this point. It needs a domestic reference procedure for how UK public bodies verify and dispute post-quantum attested proofs, including who may issue them. Without a domestic reference, UK positions in multilateral meetings lack an object other states can copy. 13. Alterations that would increase influence are procedural rather than expansive: publish a verification checklist; require fail-closed behaviour for reused or altered artefacts in any UK pilot; and keep offline verification available so checking does not depend on a live foreign service. Question 17: AI in autonomous weapons systems and related multilateral initiatives 14. AffixIO does not assess the chances of success of specific weapons-system negotiations. AffixIO notes only that debates on ethics, international law and military advantage all eventually ask a verification question: what evidence can be checked, by whom, and under what dispute process. 15. Presence and action attestation is relevant to that verification question. It is not a substitute for arms-control agreement. AffixIO claims no operational military deployment in this submission. Question 19: Lessons from earlier international regulation of technology 16. From nuclear and related verification practice, the transferable lesson is that cooperation tracks the existence of inspectable procedures more closely than the existence of aspirational principles. From passport and identity practice, the transferable lesson is that issuance authority and dispute routes must be named before a document becomes internationally usable. 17. Applied to AI: principles on model behaviour will remain contested. A shared method for verifying a machine-issued attested claim can still be agreed, tested and disputed. That is the narrow lesson AffixIO asks the Committee to take from earlier technology regimes. Recommendations 18. Ask the Ministry of Defence to assess zero-knowledge, post-quantum attestation as part of its post-quantum cryptography migration. 19. Run one non-classified pilot: attested handover. 20. Include attestation in AI governance thinking: who can issue a proof, and who can dispute it. Annex A: The 60-second test 21. This artefact is a demonstrator. It proves the yes/no relation on the bound labels below, not a live GPS fix or operational handover. Proof id: afx_a6413c475774ee7fddf076145a123a6b. Proof digest (SHA-256 hex): 13879a8e915516fa4254951ffaddc8a5c430f877d4b7316c4b0ffb9df43dc767. Circuit id: simple_yesno. Decision: yes. Generated at (UTC): 2026-11-20T07:44:36.147Z. SDK: affixio@2.0.5. 22. Example used for the demonstrator: a driver hands over a sealed box at a checkpoint. The labels bound into the artefact are Checkpoint X, time label T, and credential label Y. 23. Comparison for that example: 24. Measured size of the UltraHonk proof artefact generated for this submission: 48,492 bytes. Attestation on the accompanying envelope: HMAC-SHA256 + ML-DSA-65. The envelope is post-quantum stamped with ML-DSA-65. There is no central database to query for the offline check. The sub-90-byte claim does not apply to this artefact and is withdrawn for this submission. 25. Exact measured size: 48492 bytes (96984 hexadecimal characters). The complete proof hexadecimal string is published at https://www.affix-io.com/evidence/lords-irdc-2026/proof.hex and embedded in https://www.affix-io.com/evidence/lords-irdc-2026/artefact.json. Prefix (first 64 hex characters): 7b2270726f6f66223a5b302c302c302c302c302c302c302c302c302c302c302c. Suffix (last 64 hex characters): 3030303030303030303030303030303030303030303030303030303031225d7d. 26. Verifier link and code. Public annex folder: https://www.affix-io.com/evidence/lords-irdc-2026/. Verifier script: verify-offline.mjs. Commands: npm install affixio@2.0.5 then node verify-offline.mjs. The script performs a local UltraHonk circuit check against the expected publics in artefact.json. It does not contact AffixIO servers. 27. What the verifier learns: that the proved relation holds for the bound labels Checkpoint X, time label T, credential label Y, with decision yes on circuit simple_yesno. What the verifier does not learn: the full route; the contents of the sealed box; a full location history; a personal name; a unit identity beyond credential label Y as bound in this artefact. 28. Three steps to verify offline. (1) Install Node.js 20 or newer and run npm install affixio@2.0.5. (2) Download artefact.json and verify-offline.mjs from https://www.affix-io.com/evidence/lords-irdc-2026/. (3) Run node verify-offline.mjs and read the printed circuit_valid result. Measured offline verification time when the artefact was produced: 523 milliseconds. 29. Failure cases measured on this artefact. Presenting the same proof for a different checkpoint fails: public_mismatch:action_digest. Changing the required claim fails: public_mismatch:required_claim_hash. Substituting a forged proof string fails: malformed_proof. 30. We invite the Committee to attempt to reuse, forge, or replay this proof. Contact AffixIO hello@affix-io.com https://www.affix-io.com/ Annex materials: https://www.affix-io.com/evidence/lords-irdc-2026/ Point of comparison | Current system | AffixIO proof What is revealed | Often a full log extract or database record set sufficient to satisfy the checker | The specific handover being proved (Checkpoint X, time T, credential Y). Not the full route, contents, full location history, name, or unit How it is checked | Read the log or query a central database, usually online and account-gated | Offline local verification. Measured check time for the annex artefact: 523 milliseconds. No internet, no AffixIO account, and no call to AffixIO during the check What happens if reused or altered | A copied log extract or database screenshot can be shown again for a second event unless process controls stop it | Re-checking the same proof against a different checkpoint fails (public_mismatch:action_digest). Changing the required claim fails (public_mismatch:required_claim_hash). A forged proof string fails (malformed_proof)