CAIN-42 CAIN StudioCAIN-42 evidence library
CAIN-42 Evolution 32 -- Governed Autonomy Learning Fabric
Evidence bundle for evolution E32: 6 claims, 303 of 303 invariants held.
Last reviewed 2026-10-01
Browse the raw bundle · How to reproduce it · SHA-256 manifest
Claims (6)
| Claim | Level |
|---|---|
| C42-E32-LOOP: a 17-step governed learning loop over real governed actions: failures are mined, candidate restrict-only governance changes are compiled, sandboxed, compared on a hidden set, self-played against, canaried and promoted only through a signed gate with a registered human approval; simulated harm on a fresh workload fell from 41 to 3 | SIMULATED |
| C42-E32-NO-AUTHORITY: no learning path widens authority: every promoted configuration only restricts, drift is re-derived independently, and forged expansion signatures are refused | TESTED |
| C42-E32-WORLD-MODEL: world models learn only from direct observations, are calibrated, lose predictive privilege on collapse and are repaired through a governed ten-step path | TESTED |
| C42-E32-BENCH: 1013 of 1013 scenarios across 22 categories held; 303 of 303 invariants; mutation 12 of 12 | TESTED |
| C42-E32-SWAP: a model or runtime swap never inherits authority; only an explicit sponsor re-authorization restores action | TESTED |
| C42-E32-SCALE: synthetic in-process scale runs; the largest rows learn without execution or hash only | SIMULATED |
Invariants: 303 of 303 held
Rules the code must never break, each checked across many scenarios. See all 303 on one page.
| Niche | Held |
|---|---|
| Autonomy, control loops & recovery | 66 of 66 |
| Policy, law & governance | 54 of 54 |
| Trust & reputation | 37 of 37 |
| Identity, authority & delegation | 34 of 34 |
| Memory, data & privacy | 23 of 23 |
| Tools, MCP, protocols & adapters | 22 of 22 |
| Benchmarks, coverage & performance | 22 of 22 |
| Prediction, world models & simulation | 20 of 20 |
| Evidence, receipts & proofs | 12 of 12 |
| Core guarantees | 7 of 7 |
| Supply chain, registry & lifecycle | 5 of 5 |
| Attacks, threats & containment | 1 of 1 |
Verify it in your browser
Your browser downloads the bundle's SHA256SUMS manifest and the file(s) behind this page, hashes them with SHA-256 locally (WebCrypto), and compares. A match shows the record you are reading is the published one; it does not by itself prove who published it (see the signed claims registry and the bundle verifier for that).
Known limitations
- An in-process TESTED library; not hosted; not wired into the gateway, MCPGate or the clusters.
- Harm labels come from a SIMULATED oracle over a synthetic workload with four attack families; nothing has been learned from production traffic, and the measured harm reduction is a simulation result.
- What learning may change is a restrict-only overlay (deny prefixes, injection markers, secret patterns, a rate limit, a world-model gate); learning cannot change E30/E28/E25/E8 policy or authority.
- The world models are small statistical learners (naive Bayes and Beta-Bernoulli) over action features; they are not neural world models and do not model physics.
- The hidden evaluation is a committed synthetic workload; it is hidden from the learning code in this process, not from an adversary with access to the host.
- The governance scientist is a bounded rule-based component, not an autonomous AI researcher.
- Research ingestion is implemented but holds no external sources.
- Human approval is one registered reviewer key in the same process.
- Scale rows are synthetic and in-process; the 100,000-episode row runs learning without execution and the 1,000,000 row is hashing only.
- Hardware attestation UNKNOWN; no external review.
The bundle's own README
CAIN-42 Evolution 32 -- Governed Autonomy Learning Fabric#
CAIN learns from governed actions and may propose, test, compare, reject and roll back governance changes; it can never grant itself authority. Only restrict-only overlay configurations can be promoted, through a signed gate with a registered human approval. Harm labels are SIMULATED. Verify with verify_e32.py.txt (see REPRODUCTION.md). Status: TESTED library, pre-production.
Try CAIN-42 on your own agents
Create a free account and every new account starts with a 7-day trial of the full platform. Or try the sandbox first, with no account at all.
Create a free account → · Try the sandbox · See the whole ecosystem