White paper WP-003
The Proof-Not-Log Paradigm for AI Audit Trails
Logs tell you what someone said happened. Proofs show the maths.
At a glance
- Paper
- WP-003
- Topic
- The Proof-Not-Log Paradigm for AI Audit Trails
- Format
- PDF + web summary
- Signatures
- ML-DSA-65 post-quantum (NIST FIPS 204)
- Sandbox
- Reproducible at affix-io.com/sandbox
- Company
- AffixIO, Wales, UK
The Proof-Not-Log Paradigm for AI Audit Trails is an AffixIO technical paper. Logs tell you what someone said happened. Proofs show the maths.
Application logs are cheap and editable. For AI decisions that may be disputed years later, that is a problem. We argue for proof-not-log: ZK proofs in Merkle trees, signed with ML-DSA-65, as the primary governance artefact at a production deployment.
Summary
Application logs are cheap and editable. For AI decisions that may be disputed years later, that is a problem. We argue for proof-not-log: ZK proofs in Merkle trees, signed with ML-DSA-65, as the primary governance artefact at a production deployment.
Download the full PDF for technical detail, diagrams, and reproduction steps. Public sandbox: affix-io.com/sandbox.
Related reading
Frequently asked questions
Why are audit logs weak evidence?
Anyone with database access can alter rows; third parties cannot verify integrity without trusting the operator.
What replaces a log entry?
A proof digest, Merkle inclusion path, and signed root that any party verifies with published keys.
Can we migrate gradually?
Yes. Run proofs parallel to logs, then anchor historical log hashes into the Merkle tree before cutover.