CAIN Authority Mesh (2026-10-08)

NOT JUST "CAN CAIN STOP AN UNAUTHORIZED CALL?" BUT "CAN CAIN PROVE AUTHORITY STAYED BOUNDED ACROSS THE WHOLE CHAIN?"

What was built

An Authority Graph (identity, capability, authority, trust, reputation, provenance, policy and consent kept distinct), an Authority Lineage, an Authority Prover that returns PASS/FAIL/INCONCLUSIVE and never turns INCONCLUSIVE into PASS, multidimensional Authority Budgets, a non-broadening Policy Compiler 2.0, Policy Differential Verification, a Cryptographic Authority Receipt, an Agent Proof Graph, a Provenance Continuity engine, an Agent Immune System, a Collusion Graph, authorized deployment modes, and an Independent Authority Verifier that imports no enforcement code.

The Prompt 7 fail-open is closed

platform-gateway/cain_proof_carrying_action.py accepted unsigned or crypto-less tokens (and had an unlocked replay cache). It now fails closed with an exact verdict, supports pinned issuers and an algorithm allowlist, claims the nonce atomically, and emits a signed decision receipt for every outcome.

What was measured

• Authority adversarial swarm: 26000 attempts, 26000 blocked, 0 breaches across 13 chains.
• Governance mutation suite: 15 of 15 injected defects killed; survivors [].
• Authority Prover latency (local): p50 0.0349 ms.
• Tests: 20 authority + 9 PCAA fail-closed + 52 Sovereign + 147 Omniscience + 60 Genesis + 127 governed-evolution pass.

Reproduce: python3 scripts/authority/run_evidence.py --trials 2000 --seed 20261008.

Files: RESULTS.json · SWARM.json · MUTATION.json · LATENCY.json · manifest.json.

Limits. Libraries; the Sovereign runtime stage is still opt-in and off by default. No staging soak, no persistent stores, no non-MCP adapter, no third-party verification, no customers, no revenue, no patents. Physical AI is simulated only. PRE-PRODUCTION.