CAIN-42 CAIN Studio

Evidence library · niche

Policy, law & governance

The rules agents must follow, and how they are enforced. 966 tested invariants.

Last reviewed 2026-10-01

966 of 966 held

Where this niche's rules come from

Test families in this niche

law (642) · conformance (65) · laws (36) · policy_distribution (32) · governance_cloud (24) · obligation (19) · governance_os (18) · enforcement_binding (13) · out_of_band (8) · policy_hierarchy (8) · policy_translation (8) · meta_governance (8) · enforcement_layer (5) · moat (5) · ip (5) · scientist (4) · research_to_implementation (2) · environment_query (2) · swarm (2) · action (2) · state_machine (1) · risk (1) · translation (1) · reassessment (1) · integrity (1) · swe (1) · telemetry (1) · consistency (1) · soc (1) · operation_state (1) · gap_detector (1) · fabric_component (1) · profile (1) · lab (1) · match (1) · policy_compiler (1) · f2t (1) · graph (1)

Where these come from

Rules 151–300 of 966

IDRuleBundleResult
E28-I3expired authority does not resurrectE28held
E28-I4revoked identity does not regain authority through replayE28held
E28-I5model substitution triggers explicit continuity evaluationE28held
E28-I6runtime substitution triggers explicit continuity evaluationE28held
E28-I7capability expansion does not occur through identity migrationE28held
E28-I8consensus does not create authorizationE28held
E28-I9confidence does not create authorizationE28held
E28-I10research evidence does not automatically create authorityE28held
E28-I11simulation does not become execution authorizationE28held
E28-I12memory does not mint authority (restoring memory never restores an old lease)E28held
E28-I13a child agent cannot create authority for its parentE28held
E28-I14a subagent inherits bounded authorityE28held
E28-I15unknown remains unknown (fork telemetry, risk, hardware, translation)E28held
E28-I16a governance receipt is not forgeable without the signing keyE28held
E28-I17an execution identity is bound to its intended transactionE28held
E28-I18a stale authorization does not authorize materially changed executionE28held
E28-I19a forked identity does not silently merge with the originalE28held
E28-I20every consequential governed execution reaches the E8 Action Commit boundaryE28held
E28-I89the conformance runner executes every dimension against the real fabricE28held
E29-L1no governed identity, no governed transactionE29held
E29-L2no authority, no authorizationE29held
E29-L3no authorization, no executionE29held
E29-L4no evidence, no trust upgradeE29held
E29-L5no portable authorityE29held
E29-L6unknown never becomes verifiedE29held
E29-L7agreement is not authorizationE29held
E29-L8reputation is not authorizationE29held
E29-L9discovery is not trustE29held
E29-L10authority to act is not authority to complete every transactionE29held
E29-L11no money moves without an E8-committed governance receiptE29held
E29-L12a claim of success never releases escrowE29held
E29-L13confidence never resolves a disputeE29held
E29-L14degraded mode never escalates privilegeE29held
E29-L15organizational scale never creates unbounded authorityE29held
E29-L16no agent promotes its own code to productionE29held
E29-L17a learned rule never activates without a governed approvalE29held
E29-L18a restart never restores authorityE29held
E29-L19graph intelligence is a risk signal, never a grantE29held
E29-L20routing never bypasses E8E29held
E30-L1no verified identity, no trustE30held
E30-L2no authority, no authorizationE30held
E30-L3no authorization, no executionE30held
E30-L4no enforcement, no claim of controlE30held
E30-L5no evidence, no trust upgradeE30held
E30-L6unknown never becomes verifiedE30held
E30-L7perception is not reality; confidence is not truthE30held
E30-L8memory never mints authorityE30held
E30-L9reasoning is not authority; more compute is not more authorityE30held
E30-L10consensus is not authorizationE30held
E30-L11self-improvement never bypasses governanceE30held
E30-L12self-healing may propose a repair, never authorize its own deploymentE30held
E30-L13model output is not world actionE30held
E30-L14no agent promotes its own autonomy levelE30held
E30-L15REVOKED never goes straight to AUTHORIZED; UNKNOWN never becomes TRUSTEDE30held
E30-L16every consequential action consumes autonomy budget; replenishment is governedE30held
E30-L17crossing a containment boundary needs a new authorizationE30held
E30-L18monitored is not enforcedE30held
E30-L19screenshots and visual observations are evidence, not authorityE30held
E30-L20knowledge is superseded or revoked, never silently deletedE30held
E30-H1self-healing deploys only with two registered humansE30held
E31-L1no identity, no governance proofE31held
E31-L2no authority, no governance proofE31held
E31-L3no authorization, no governed executionE31held
E31-L4no enforcement, no claim of controlE31held
E31-L5no evidence, no verified outcomeE31held
E31-L6no proof, no verified governance claimE31held
E31-L7a governance proof never creates authorityE31held
E31-L8a receipt never becomes permissionE31held
E31-L9a proof of past authorization never authorizes a future actionE31held
E31-L10a valid proof for action A never authorizes action BE31held
E31-L11a child agent cannot prove more authority than its parentE31held
E31-L12delegation is traceableE31held
E31-L13revocation propagatesE31held
E31-L14expired authority does not resurrectE31held
E31-L15replayed proofs do not authorize executionE31held
E31-L16stale proofs do not authorize executionE31held
E31-L17altered proofs fail verificationE31held
E31-L18unknown remains unknownE31held
E31-L19simulation never becomes reality evidenceE31held
E31-L20prediction never becomes authorityE31held
E31-L21model confidence never creates authorityE31held
E31-L22consensus never creates authorityE31held
E31-L23reputation never creates authorityE31held
E31-L24economic value never creates authorityE31held
E31-L25a proof describes what was verified, not what was assumedE31held
E31-L26outside an enforcement boundary CAIN does not claim controlE31held
E31-L27a verifier can reject a false governance claimE31held
E31-L28a governance proof is bound to the specific actionE31held
E31-L29every consequential governed action is traceable to E8E31held
E31-L30evidence remembers the governance decisionE31held
E31-LAYER-DECISIONDECISION is UNKNOWN when its artifact is missing, never VERIFIEDE31held
E31-LAYER-AUTHORIZATIONAUTHORIZATION is UNKNOWN when its artifact is missing, never VERIFIEDE31held
E31-LAYER-ENFORCEMENTENFORCEMENT is UNKNOWN when its artifact is missing, never VERIFIEDE31held
E31-LAYER-EXECUTIONEXECUTION is UNKNOWN when its artifact is missing, never VERIFIEDE31held
E31-LAYER-OUTCOMEOUTCOME is UNKNOWN when its artifact is missing, never VERIFIEDE31held
E31-CF-G0G0 (identity observed) cannot be skippedE31held
E31-CF-G1G1 (identity verified) cannot be skippedE31held
E31-CF-G2G2 (authority verified) cannot be skippedE31held
E31-CF-G3G3 (authorization verified) cannot be skippedE31held
E31-CF-G4G4 (execution boundary verified) cannot be skippedE31held
E31-CF-G5G5 (proof-carrying execution) cannot be skippedE31held
E31-CF-G6G6 (independent proof verification) cannot be skippedE31held
E31-CF-G7G7 (continuous governance and revocation verification) cannot be skippedE31held
E31-CF-G8G8 (end-to-end governed execution with verified postconditions) cannot be skippedE31held
E31-TC-identicala identical policy translation is classified correctlyE31held
E31-TC-reordered_duplicateda reordered_duplicated policy translation is classified correctlyE31held
E31-TC-redundant_rulea redundant_rule policy translation is classified correctlyE31held
E31-TC-narroweda narrowed policy translation is classified correctlyE31held
E31-TC-wideneda widened policy translation is classified correctlyE31held
E31-TC-swappeda swapped policy translation is classified correctlyE31held
E31-TC-unenforceablea unenforceable policy translation is classified correctlyE31held
E31-TC-unknown_controla unknown_control policy translation is classified correctlyE31held
E31-EB-drop_e8enforcement binding holds against drop_e8E31held
E31-EB-swap_e8enforcement binding holds against swap_e8E31held
E31-EB-e8_denied_rehashedenforcement binding holds against e8_denied_rehashedE31held
E31-EB-e25_not_executedenforcement binding holds against e25_not_executedE31held
E31-EB-forged_issuer_layersenforcement binding holds against forged_issuer_layersE31held
E31-EB-insider_proof_over_denialenforcement binding holds against insider_proof_over_denialE31held
E31-EB-logged_not_enforcedenforcement binding holds against logged_not_enforcedE31held
E31-EB-e8_action_rehashedenforcement binding holds against e8_action_rehashedE31held
E31-EB-swap_e28enforcement binding holds against swap_e28E31held
E31-EB-swap_e25enforcement binding holds against swap_e25E31held
E31-EB-swap_uarenforcement binding holds against swap_uarE31held
E31-EB-e8_from_other_agentenforcement binding holds against e8_from_other_agentE31held
E31-EB-attacker_anchorsenforcement binding holds against attacker_anchorsE31held
E32-L1learning never creates authorityE32held
E32-L2experience never becomes authorityE32held
E32-L3prediction never becomes factE32held
E32-L4simulation never becomes reality evidenceE32held
E32-L5model confidence never creates permissionE32held
E32-L6reward never creates authorityE32held
E32-L7success never proves safetyE32held
E32-L8failure becomes evidenceE32held
E32-L9unknown remains unknownE32held
E32-L10memory does not become truth merely by persistingE32held
E32-L11repeated observations do not automatically become certaintyE32held
E32-L12self-improvement does not self-authorizeE32held
E32-L13a learned policy does not automatically replace a governed policyE32held
E32-L14a new model does not inherit authority without re-evaluationE32held
E32-L15a new runtime does not inherit authority without continuity verificationE32held
E32-L16a new skill does not inherit authority automaticallyE32held
E32-L17a new capability does not inherit authority automaticallyE32held
E32-L18a new agent does not inherit parent authority without attenuated delegationE32held
E32-L19a better benchmark score does not bypass governanceE32held
E32-L20ensemble agreement does not create authorityE32held
E32-L21world-model agreement does not create authorityE32held
E32-L22research consensus does not create authorityE32held
E32-L23collective agent consensus does not create authorityE32held
E32-L24economic value does not create authorityE32held
E32-L25reputation does not create authorityE32held

1 2 3 4 5 6 7

Other niches

Consensus & distributed systems · Attacks, threats & containment · Identity, authority & delegation · Evidence, receipts & proofs · Memory, data & privacy · Prediction, world models & simulation · Transactions, markets & economics · Tools, MCP, protocols & adapters · Autonomy, control loops & recovery · Trust & reputation · Supply chain, registry & lifecycle · Benchmarks, coverage & performance · Core guarantees

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