Demo AI Platform — EU AI Act Preview

Found {factor_b} active safety guardrails (required: {factor_a}). EU AI Act Art. 9(2)(a) requires suitable risk management measures to identify and mitigate risks. NIST AI RMF MANAGE 4.1 requires risk treatments. Guardrail enforcement: {verdict_word}met.

Open 81d

Overseer identity capture is {verdict_word}active. EU AI Act Art. 12(2)(d) requires logs to include identification of natural persons involved in verification of results. NIST AI RMF GOVERN 1.1 requires traceable oversight.

Open 177d (overdue)

Explanation generation is {verdict_word}active. EU AI Act Art. 13(1) requires high-risk AI systems to be sufficiently transparent to enable deployers to interpret output. NIST AI RMF MEASURE 2.5 requires interpretability measures.

Open 175d (overdue)

Content safety filter {verdict_word}passed. EU AI Act Art. 9(4)(b) requires risk management to address reasonably foreseeable misuse including harmful outputs. NIST AI RMF GOVERN 1.5 requires ongoing policy compliance assessment.

Open 174d (overdue)

Fairness calibration score is {factor_b} against a minimum of {factor_a}. EU AI Act Art. 9(4)(a) requires elimination or reduction of risks. NIST AI RMF MAP 2.3 requires mapping AI impacts to affected populations. Score {verdict_word}meets minimum.

Open 173d (overdue)

Model confidence score is {factor_b} against a minimum of {factor_a}. EU AI Act Art. 13(3)(b)(ii) requires disclosure of the degree of accuracy. NIST AI RMF MAP 2.3 requires confidence calibration. Score {verdict_word}meets threshold.

Open 172d (overdue)

Approved model registry is {verdict_word}current. EU AI Act Art. 49 requires registration of high-risk AI systems in the EU database. NIST 800-53 CM-8 requires a system component inventory including version and risk classification.

Open 171d (overdue)

Emergency override is {verdict_word}authorized: trigger type {factor_a} (0=e-stop, 1=operator, 2=escalation, 3=responder), authorization level {factor_b} (0=operator, 1=supervisor, 2=manager, 3=responder), fallback state {factor_c} (0=safe, 1=legacy, 2=manual, 3=degraded, 4=shutdown). EU AI Act Art. 14 requires human override capability for high-risk AI. IEC 61511 requires logged, auditable override events.

Open 81d

Drift consequence response is {verdict_word}proportional: drift magnitude {factor_a}, consequence category {factor_b} (0=safety, 1=environmental, 2=financial, 3=operational, 4=reputational), response action {factor_c} (0=notify, 1=monitor, 2=throttle, 3=circuit-break, 4=failover, 5=shutdown). EU AI Act Art. 9(2)(b) requires continuous risk estimation. OCC 2026-13 requires materiality-mapped drift tracking with graduated responses.

Open 81d

Champion-challenger assessment is {verdict_word}within acceptance criteria: {factor_a} inputs processed, max divergence {factor_b}/1000, threshold breached = {factor_c} (1=yes, 0=no). EU AI Act Art. 15 requires post-market monitoring. OCC 2026-13 requires versioned challenger runs with documented sign-off.

MCP tool integrity is {verdict_word}intact: schema hash {factor_a}, invocation #{factor_b}, drift detected = {factor_c} (0=none, 1=drift). OWASP Agentic MCP-03 (Tool Poisoning) requires schema integrity monitoring between invocations.

Hazard reachability is {verdict_word}mitigated: {factor_b} of {factor_a} assessed hazards confirmed mitigated, method {factor_c} (0=manual, 1=automated, 2=vendor_attested). EU AI Act Art. 53 requires risk identification. ISO 42001 6.1.2 requires risk assessment.

MCP server authentication is {verdict_word}active: method {factor_a} (0=none, 1=api_key, 2=oauth, 3=mtls, 4=did), validity {factor_b}s, mutual auth = {factor_c}. OWASP Agentic MCP-07 requires server authentication. NIST 800-53 IA-9 requires service identification.

Training data statistics are {verdict_word}recorded: {factor_a} total samples, {factor_b} features, class balance ratio = {factor_c} (x1000, e.g., 850 = 85%). EU AI Act Art. 10(3) requires documentation of training data properties and statistical characteristics.

Training data PII lifecycle event is {verdict_word}recorded: {factor_a} records affected, event completed = {factor_b} (1=yes, 0=partial/failed), event type code {factor_c} (0=unspecified, 1=pseudonymization, 2=anonymization, 3=access_restriction, 4=deletion, 5=encryption). EU AI Act Art. 10(5) requires data governance. GDPR Art. 25 requires data protection by design.

MCP server discovery is {verdict_word}complete: method {factor_a} (0=manual, 1=dns-sd, 2=mdns, 3=registry, 4=network), {factor_b} servers found, {factor_c} unauthorized (shadow servers). OWASP Agentic MCP-09 requires authorized server inventory. NIST 800-53 CM-8 requires component inventory.

AI risk management scope definition is {verdict_word}documented. NIST AI RMF GOVERN 1.3 requires a documented scope definition including which AI systems are in scope, their intended purposes, deployment contexts, and risk tolerances. NIST 800-53 SA-3 requires system development lifecycle planning that defines scope boundaries.

AI governance resource allocation is {verdict_word}documented. NIST AI RMF GOVERN 2.2 requires adequate resources (budget, staffing, compute) allocated for AI risk management activities. NIST 800-53 PM-3 requires information security resource planning.

AI risk identification and categorization is {verdict_word}complete. NIST AI RMF MAP 2.1 requires identified risk sources with assigned categories and documented residual risk acceptance decisions. NIST 800-53 RA-2 requires security categorization of information systems including AI components.

AI incident response capability is {verdict_word}operational. NIST AI RMF MANAGE 3.1 requires real-time incident detection and response procedures for AI system failures. NIST 800-53 IR-8 requires an incident response plan covering AI-specific scenarios including model recall, guardrail bypass, and data contamination.

AI societal impact assessment is {verdict_word}complete. NIST AI RMF MAP 5.2 requires community and societal-level impact assessment beyond individual bias metrics. NIST 800-53 RA-3 requires risk assessment that considers broader organizational and societal impacts.

RAG context relevance is {verdict_word}assessed: similarity threshold {factor_a}/1000, average similarity {factor_b}/1000, {factor_c} chunks below threshold. EU AI Act Art. 10(2) requires data governance covering data quality and relevance. NIST AI RMF MEASURE 2.6 requires traceability of AI system outputs.

Model weight file integrity is {verdict_word}verified: hash match = {factor_b} (1=match, 0=mismatch). EU AI Act Art. 15(4) requires resilience against unauthorized modification. NIST AI RMF MANAGE 1.3 requires model integrity verification.

Adapter stack is {verdict_word}attested: {factor_a} adapters loaded, all verified = {factor_b} (1=yes, 0=no). EU AI Act Art. 12(2)(b) requires version and lineage tracking. NIST AI RMF MAP 2.3 requires documentation of AI system components.

Model quantization is {verdict_word}attested: method code {factor_c} (0=FP32, 1=FP16, 2=BF16, 3=INT8, 4=INT4, 5=GPTQ, 6=AWQ, 7=GGUF). EU AI Act Art. 15(3) requires performance consistency documentation. NIST AI RMF MEASURE 2.5 requires computational accuracy assessment.

Memory context is {verdict_word}bound: {factor_a} memory sources active, all identified = {factor_b} (1=yes, 0=anonymous). EU AI Act Art. 12(2)(a) requires logging of reference data sources. NIST AI RMF MAP 3.5 requires data provenance documentation.

Reward model is {verdict_word}bound: identified = {factor_b} (1=yes, 0=unknown). EU AI Act Art. 9(4)(a) requires risk identification including alignment mechanisms. NIST AI RMF MEASURE 2.6 requires traceability of output scoring.

TPM platform integrity is {verdict_word}attested: {factor_a} PCR registers read, all non-zero = {factor_b} (1=healthy, 0=uninitialized/tampered), reserved {factor_c}. Proves host firmware has not been tampered with via hardware root of trust. NIST 800-53 SC-12 requires cryptographic key establishment. EU AI Act Art. 15(4) requires cybersecurity measures.

Trust handshake is {verdict_word}recorded: {factor_a} checks performed, {factor_b} passed, result {factor_c} (1=granted, 0=denied). Bilateral evidence of agent-to-agent trust verification. EU AI Act Art. 12 requires record-keeping for inter-system interactions. NIST AI RMF GOVERN 1.4 requires risk-proportional access controls.

Chain trust degradation is {verdict_word}recorded: previous trust = {factor_a}, new trust = {factor_b}, degradation delta = {factor_c}. EU AI Act Art. 9 requires risk management through multi-agent chains. NIST AI RMF GOVERN 1.3 requires trust level tracking across handoffs.

Bias audit is {verdict_word}recorded: {factor_a} demographic groups tested, {factor_b} disparities found, max disparity = {factor_c} (x10, e.g., 125 = 12.5%). EU AI Act Art. 10(2)(f) requires examination of bias. NIST AI RMF MAP 2.3 requires bias identification across demographic groups.

Dependency manifest is {verdict_word}attested: {factor_a} dependencies counted, all pinned = {factor_b} (1=yes, 0=no), {factor_c} known vulnerabilities. EU AI Act Art. 11 requires technical documentation. EO 14028 requires software supply chain integrity.

Agent behavioral baseline is {verdict_word}recorded: {factor_a} behavioral dimensions measured, within envelope = {factor_b} (1=yes, 0=drift), mode {factor_c} (0=establishing, 1=monitoring, 2=drift_detected, 3=baseline_reset). EU AI Act Art. 9(2)(b) requires continuous monitoring of AI system performance. NIST AI RMF MEASURE 2.6 requires measurement of AI system trustworthiness over time.

License provenance is {verdict_word}verified: {factor_a} license components checked, all compliant = {factor_b} (1=yes, 0=violation detected), license type code {factor_c} (0=permissive, 1=copyleft, 2=proprietary, 3=dual, 4=openmdw, 5=unknown). EU AI Act Art. 53(1)(d) requires transparency of training data and model licensing for GPAI providers. NIST AI RMF GOVERN 1.7 requires documenting legal and licensing constraints.

Adversarial test campaign is {verdict_word}recorded: {factor_a} attack scenarios executed, {factor_b} successfully mitigated, coverage category code {factor_c} (0=prompt_injection, 1=jailbreak, 2=data_poisoning, 3=model_extraction, 4=membership_inference, 5=adversarial_examples, 6=supply_chain, 7=denial_of_service, 8=output_manipulation, 9=privilege_escalation, 10=comprehensive). EO 14110 requires red-team testing of foundation models. EU AI Act Art. 9(7) requires testing including adversarial testing.

Multi-agent delegation is {verdict_word}recorded: {factor_a} delegation depth (hops from human), {factor_b} permissions granted, time bound {factor_c} minutes (0=unbounded). EU AI Act Art. 9 requires risk management for autonomous delegation chains. NIST AI RMF GOVERN 1.3 requires documenting roles and responsibilities across agent boundaries.

Incident report is {verdict_word}recorded: severity code {factor_a} (1=low, 2=medium, 3=high, 4=critical), authority notified = {factor_b} (1=yes, 0=no), incident type code {factor_c} (0=safety, 1=rights, 2=security, 3=performance, 4=bias, 5=other). EU AI Act Art. 62 requires reporting serious incidents. NIST AI RMF MANAGE 3.2 requires incident communication.

Performance metrics are {verdict_word}recorded: {factor_a} metrics evaluated, {factor_b} passing, benchmark type code {factor_c} (0=accuracy, 1=precision, 2=recall, 3=f1, 4=auc, 5=custom). EU AI Act Art. 15(1) requires declared accuracy levels. NIST AI RMF MEASURE 2.5 requires performance measurement.

Robustness testing is {verdict_word}recorded: {factor_a} perturbations tested, {factor_b} survived, perturbation type code {factor_c} (0=noise, 1=corruption, 2=missing_data, 3=out_of_distribution, 4=edge_case, 5=adversarial_input). EU AI Act Art. 15(3) requires resilience against errors and inconsistencies. NIST AI RMF MEASURE 2.6.

Cybersecurity assessment is {verdict_word}recorded: {factor_a} controls assessed, {factor_b} compliant, framework code {factor_c} (0=nist_csf, 1=iso27001, 2=owasp, 3=cis, 4=custom). EU AI Act Art. 15(4) requires appropriate cybersecurity. NIST Cybersecurity Framework.

Automated decision notification is {verdict_word}recorded: {factor_a} decisions made, {factor_b} human reviewed, decision type code {factor_c} (0=credit, 1=employment, 2=insurance, 3=benefits, 4=legal, 5=other). GDPR Art. 22 requires notification of automated decisions with legal effects. EU AI Act Art. 14 requires human oversight.

Dual-use model classification is {verdict_word}recorded: classification code {factor_a} (0=standard, 1=dual_use, 2=high_impact), reporting status code {factor_b} (0=not_required, 1=pending, 2=notified, 3=acknowledged), {factor_c} days since classification. EO 14110 Sec 4(a) requires reporting dual-use foundation models. NIST AI RMF GOVERN 1.1.

Orchestration topology is {verdict_word}recorded. Topology: {factor_a} (0=sequential, 1=parallel, 2=hierarchical, 3=hybrid, 4=mesh), agents: {factor_b}, dependency depth: {factor_c}. NIST AI RMF GOVERN 1.3 requires documenting multi-agent coordination patterns. EU AI Act Art. 9 requires risk management for AI systems.

Inter-agent handoff is {verdict_word}recorded. Permission delta: {factor_c} (1=escalation, 0=lateral, -1=restriction). NIST AI RMF GOVERN 1.3 requires tracking delegation between AI components. EU AI Act Art. 9 requires documenting AI system interactions.

Context window management is {verdict_word}within bounds. Tokens before: {factor_a}, after: {factor_b}, eviction method: {factor_c} (0=none, 1=truncation, 2=summarization, 3=sliding window, 4=priority eviction). NIST AI RMF MEASURE 2.6 requires transparency of context handling. EU AI Act Art. 13 requires transparency measures.

Sandbox enforcement is {verdict_word}clean. Tools declared: {factor_a}, invoked: {factor_b}, violations: {factor_c}. NIST 800-53 SA-11(8) requires software fault isolation. EU AI Act Art. 15 requires cybersecurity measures for high-risk AI systems. This procedure witnesses the harness's own report; cross-reference AI-TOOL.1 for independent verification.

Eval gate decision is {verdict_word}above threshold. Total evals: {factor_a}, passed: {factor_b}, gate score: {factor_c}/100. NIST AI RMF MEASURE 2.5 requires documented evaluation criteria. EU AI Act Art. 9(7) requires testing appropriate to the intended purpose of the AI system.

A2A task delegation is {verdict_word}witnessed: state code {factor_a} (0=submitted, 1=working, 2=input_required, 3=completed, 4=failed, 5=canceled, 6=rejected), latency {factor_b}ms, delegation depth {factor_c}. EU AI Act Art. 9 requires risk management across AI system interactions. NIST AI RMF GOVERN 1.3 requires accountability for delegated operations.

Supply chain risk is {verdict_word}assessed: {factor_a} suppliers assessed, {factor_b} compliant, risk level code {factor_c} (0=low, 1=medium, 2=high, 3=critical). NIST AI RMF MEASURE 3.1 requires supply chain risk metrics. G7/CISA SBOM-AI. EO 14028.

Post-market monitoring is {verdict_word}recorded: {factor_a} checks run, {factor_b} anomalies detected, monitoring type code {factor_c} (0=performance, 1=fairness, 2=safety, 3=security, 4=comprehensive). EU AI Act Art. 72 requires post-market monitoring system. NIST AI RMF MANAGE 4.1.

Autonomous generation depth is {verdict_word}within limits: max allowed depth = {factor_a}, observed depth = {factor_b}, human gate present = {factor_c} (1=yes, 0=autonomous). EU AI Act Art. 14 requires human oversight proportionate to risk. NIST AI RMF GOVERN 1.5 requires organizational policies for autonomous systems. EO 14110 Sec. 3 addresses recursive AI generation risks.

External timestamp attestation is {verdict_word}verified: {factor_a} anchors in batch, TSA verified = {factor_b} (1=yes, 0=no), TSA provider code {factor_c} (0=none, 1=freetsa, 2=digicert, 3=sectigo, 4=custom). EU AI Act Art. 12(1) requires tamper-evident logging. NIST AI RMF GOVERN 1.4 requires organizational accountability. RFC 3161 timestamping provides non-repudiation.

Agent card discovery is {verdict_word}verified: method {factor_a} (0=direct_url, 1=well_known, 2=registry, 3=referral), {factor_b} agents discovered, {factor_c} with verifiable credentials. EU AI Act Art. 13 requires transparency measures. Five Eyes Agentic FE-5 requires agent identity verification.

Context chain linking is {verdict_word}recorded: {factor_a} tasks in chain, {factor_c} distinct agents participating. EU AI Act Art. 9 requires traceable risk management across multi-agent interactions. Five Eyes Agentic FE-6 requires chain-of-custody tracking.

OAuth token binding is {verdict_word}enforced: event type {factor_a} (0=discovery through 7=revocation), {factor_b} scopes, binding strength {factor_c} (0=none, 1=session, 2=dpop, 3=mtls). OWASP Agentic MCP-07 requires token binding. EU AI Act Art. 15(3) requires access control measures.

AI system decommissioning is {verdict_word}complete: {factor_b} of {factor_a} downstream dependencies resolved, disposal method {factor_c} (0=archived, 1=destroyed, 2=isolated, 3=transferred). EU AI Act Art. 9 requires lifecycle risk management. ISO 42001 8.2 requires operational planning.

Governance infrastructure attestation is {verdict_word}recorded: {factor_a} governance components attested, {factor_b} policies verified, configuration hash present = {factor_c} (1=yes, 0=no). Governance systems must attest their own configuration using the same protocol they enforce on governed systems.

Governance layer registration is {verdict_word}recorded: {factor_a} governance layers registered, {factor_b} validated against witness layer, registration status = {factor_c} (1=authoritative, 0=pending). Each AI governance layer must register with the non-AI witness layer before its outputs are considered authoritative.

SIM-bound attestation chain is {verdict_word}intact: binding method {factor_c} (0=none, 1=iccid, 2=imsi_hash, 3=esim_eid, 4=dual_sim), offline integrity = {factor_b} (1=intact, 0=broken). EU AI Act Art. 15(4) requires cybersecurity measures for edge AI. NIST AI RMF MANAGE 2.4 requires operational safeguards.

Roaming policy is {verdict_word}compliant: policy disposition {factor_c} (0=allow, 1=restrict, 2=deny, 3=escalate). EU AI Act Art. 9(7) requires risk management for AI operating across jurisdictions. NIST AI RMF GOVERN 1.1 requires governance of cross-boundary operations.

Dual-SIM data residency is {verdict_word}compliant: {factor_a} active SIM profiles, routing mode {factor_c} (0=primary_only, 1=failover, 2=load_balance, 3=geo_fenced). GDPR Art. 44-49 requires lawful cross-border data transfers. NIS-2 Art. 21 requires network security measures.

Peer-to-peer trust mesh has {factor_b} verified peers out of {factor_a} total: mesh mode {factor_c} (0=isolated, 1=relay, 2=direct, 3=broadcast). EU AI Act Art. 15(1) requires technical robustness. NIST AI RMF MANAGE 2.2 requires trustworthiness of connected systems.

Bilateral flush correlation is {verdict_word}complete: {factor_b} of {factor_a} locally buffered anchors correlated with clearing house, flush mode {factor_c} (0=manual, 1=connectivity_restored, 2=scheduled, 3=peer_relay). EU AI Act Art. 12 requires logging integrity. NIST AI RMF MEASURE 2.5 requires measurement completeness.

Policy downgrade detection is {verdict_word}recorded: {factor_a} policy transitions evaluated, {factor_b} downgrades detected, enforcement status = {factor_c} (1=blocked, 0=allowed). Governance systems must enforce monotonic policy version progression and detect unauthorized downgrades.

Circular dependency check is {verdict_word}recorded: {factor_a} governance rules evaluated, {factor_b} cycles detected, validation method = {factor_c} (1=topological_sort, 0=manual). Governance rule configurations must be validated for circular dependencies before activation.

Governance authorization is {verdict_word}recorded: {factor_a} authorization requests processed, {factor_b} approved, signature verification = {factor_c} (1=verified, 0=unsigned). Governance configuration changes require cryptographically signed authorization from operators with appropriate authority scope.

Emergency override attestation is {verdict_word}recorded: {factor_a} emergency changes made, {factor_b} reviewed within window, auto-witnessed = {factor_c} (1=yes, 0=no). Emergency governance changes made outside normal approval workflow must be automatically witnessed and trigger mandatory review.

Governance sync verification is {verdict_word}recorded: {factor_a} federated peers checked, {factor_b} policy equivalence confirmed, governance floor met = {factor_c} (1=yes, 0=divergence). Federated organizations must verify governance policy equivalence during trust credential exchange.

Attestation purity verification is {verdict_word}recorded: {factor_a} source modules hashed, {factor_b} ML-free verified, purity status = {factor_c} (1=pure, 0=contaminated). Attestation engine source code must be verified to contain no machine learning inference in the witness path.

AI system re-commissioning is {verdict_word}validated: {factor_b} of {factor_a} safety checks passed, type {factor_c} (0=full_validation, 1=shadow_mode, 2=limited_scope). EU AI Act Art. 72 requires post-incident validation. ISO 42001 10.2 requires corrective action verification.

Witnessing density is {verdict_word}sufficient: {factor_b} of {factor_a} expected anchors observed, status {factor_c} (0=sufficient, 1=insufficient, 2=degraded). EU AI Act Art. 9 requires continuous risk management. NIST AI RMF MEASURE 2.6 requires ongoing measurement.

Network slice isolation is {verdict_word}verified: slice ID hash {factor_a}, isolation status {factor_b} (0=not verified, 1=verified), cross-slice requests blocked: {factor_c}. 3GPP TS 28.310 requires network slice lifecycle management. GSMA NG.116 requires slice isolation assurance. 3GPP TS 33.501 Sec. 6.8 governs slice security.

Design generation provenance is {verdict_word}recorded: {factor_a} constraints applied, {factor_b} parameters generated, design domain code {factor_c} (0=mechanical, 1=chemical, 2=electrical, 3=structural, 4=thermal, 5=pharmaceutical, 6=semiconductor, 7=custom). DO-178C 5.1 requires traceable development process. ASME V&V 10 3.1 requires model development documentation. FDA 21 CFR 11.10(a) requires system validation.

Simulation validation is {verdict_word}recorded: {factor_a} simulations run, {factor_b} passed, simulation type code {factor_c} (0=fea, 1=cfd, 2=molecular, 3=thermal, 4=electromagnetic, 5=multiphysics, 6=custom). DO-178C 6.3 requires test coverage analysis. ASME V&V 10 4.1 requires computational model validation. ISO 26262-4 requires system verification.

Safety-critical review gate is {verdict_word}recorded: {factor_a} reviewers required, {factor_b} approved, approval type code {factor_c} (0=peer, 1=pe_stamp, 2=safety_board, 3=regulatory, 4=independent_assessor). DO-178C 7.2 requires review and analysis of outputs. FDA 21 CFR 11.10(g) requires authority checks. ISO 26262-2 requires safety management.

Material specification compliance is {verdict_word}recorded: {factor_a} specifications checked, {factor_b} met, standard code {factor_c} (0=asme, 1=iso, 2=astm, 3=mil_spec, 4=fda_usp, 5=iec, 6=custom). ASME V&V 10 3.3 requires requirements traceability. DO-254 5.3 requires requirements validation. ISO 26262-8 requires supporting processes.

Design revision chain is {verdict_word}recorded: {factor_a} total revisions, {factor_b} AI-generated, chain status code {factor_c} (0=in_progress, 1=approved, 2=rejected, 3=superseded, 4=archived). DO-178C 7.3 requires configuration management. FDA 21 CFR 11.10(e) requires audit trail. ASME V&V 10 2.4 requires documentation of model history.

Fabrication release is {verdict_word}authorized: design hash verified = {factor_a} (1=match, 0=mismatch), {factor_b} authorizations received, release type code {factor_c} (0=prototype, 1=limited_run, 2=mass_production, 3=field_modification, 4=emergency). DO-178C 5.5 requires integration verification. FDA 21 CFR 11.10(f) requires operational system checks. ISO 26262-4 7.4.4 requires production release validation.

Model parameter freeze is {verdict_word}intact: {factor_b} of {factor_a} locked configuration items verified unchanged, scope {factor_c} (0=weights_only, 1=config_and_params, 2=full_stack). EU AI Act Art. 9(2)(b) requires controls over operational drift. NIST 800-53 CM-3 requires configuration change control.

Agent transaction witnessing is {verdict_word}recorded: authorization type = {factor_a} (0=none, 1=pre-approved, 2=human, 3=policy, 4=budget_limit), amount = {factor_b} cents, status = {factor_c} (0=pending, 1=authorized, 2=denied, 3=escalated). EU AI Act Art. 14 requires human oversight of automated decisions. NIST AI RMF MANAGE 2.2 requires risk response tracking.

Agent lifecycle event is {verdict_word}recorded: event type = {factor_a} (0=spawn, 1=checkpoint, 2=migrate, 3=terminate, 4=crash), context tokens = {factor_b}, state hash present = {factor_c} (1=yes, 0=no). EU AI Act Art. 12 requires automatic recording of events. NIST AI RMF MAP 1.3 requires lifecycle stage characterization.

Trajectory decision is {verdict_word}attested: safety validated = {factor_b} (1=passed, 0=failed), classification {factor_c} (1=nominal, 2=cautionary, 3=degraded, 4=emergency, 5=abort). EU AI Act Annex III(3a) covers safety components of vehicles. ISO/PAS 8800 requires operational design domain validation.

VLA inference is {verdict_word}recorded: latency {factor_b}ms, succeeded = {factor_c} (1=yes, 0=exception/timeout). EU AI Act Art. 12(1) requires automatic logging of AI system operation. ISO/PAS 8800 requires real-time inference monitoring for autonomous systems.

Tool permission attestation is {verdict_word}recorded: {factor_a} tools granted, charter match = {factor_b} (1=matches, 0=drift), permission change type = {factor_c} (0=none, 1=added, 2=removed, 3=escalated). EU AI Act Art. 9 requires risk management. NIST AI RMF GOVERN 1.5 requires ongoing monitoring of AI system capabilities.

Cross-border inference routing is {verdict_word}recorded: serving region = {factor_a} (ISO 3166 numeric), user region = {factor_b} (ISO 3166 numeric), compliance status = {factor_c} (0=unchecked, 1=compliant, 2=blocked, 3=override). EU AI Act Art. 10(5) requires data governance for cross-border processing. NIST AI RMF GOVERN 1.7 requires context-specific risk management.

Probabilistic witnessing summary recorded: {factor_a} inferences skipped during the sampling window. Summary anchors always PASS. EU AI Act Art. 12 requires logging completeness. NIST AI RMF MEASURE 2.6 requires measurement scope documentation.

MCP security posture is {verdict_word}sufficient: {factor_b} of {factor_a} checks passed, score {factor_c}/100 (minimum: 75). NSA/CSA MCP Security Best Practices requires observable security posture. OWASP Agentic Top 10 covers the full threat surface.

Model provenance chain is {verdict_word}verified: {factor_a} links in lineage, integrity = {factor_b} (1=verified, 0=unverified), link type {factor_c} (0=training, 1=fine_tuning, 2=deployment, 3=distillation). EU AI Act Art. 11 requires technical documentation. NIST AI RMF MAP 1.1 requires data provenance.

Delegation boundary is {verdict_word}enforced: max depth {factor_a}, actual depth {factor_b}, boundary action {factor_c} (0=blocked, 1=warned, 2=escalated, 3=allowed). EU AI Act Art. 14 requires human oversight. NIST AI RMF GOVERN 1.4 requires delegation governance.

Model registry integrity is {verdict_word}verified: {factor_b} of {factor_a} expected model versions confirmed present and checksum-validated. EU AI Act Art. 11 requires technical documentation. NIST 800-53 CM-3 requires configuration change control.

Model version recording is {verdict_word}active. EU AI Act Art. 12(2)(b) requires logs to include identification of the reference database. NIST AI RMF MANAGE 2.2 requires post-deployment version monitoring.

AI inference log retention is {factor_b} days (minimum required: {factor_a} days). EU AI Act Art. 12(3) requires deployers to keep automatically generated logs for at least six months. Retention {verdict_word}compliant.

Feedback loop isolation is {verdict_word}active. EU AI Act Art. 15(4) requires controls to prevent biased outputs from contaminating training data in systems that continue learning after deployment. NIST AI RMF MANAGE 2.2 requires ongoing risk tracking.

PII redaction is {verdict_word}active. EU AI Act Art. 10(2)(f) requires examination of possible biases affecting health and safety, with data privacy measures. NIST AI RMF GOVERN 1.7 requires data protection processes.

Bias disparity ratio is {factor_b} against a maximum of {factor_a}. EU AI Act Art. 10(2)(f) requires examination of possible biases in training datasets and outputs. NIST AI RMF MEASURE 2.5 requires bias evaluation across demographic groups. Disparity {verdict_word}within threshold.

Training data provenance is {verdict_word}documented. EU AI Act Art. 10(2)(a) requires documenting relevant design choices including data collection processes. NIST AI RMF MAP 3.5 requires documentation of data provenance and lineage.

Training data license compliance is {verdict_word}verified. EU AI Act Art. 10(2)(a) requires data governance covering design choices and data origin. NIST AI RMF GOVERN 1.7 requires conformance with legal and regulatory data requirements.

Human override event logging is {verdict_word}active. EU AI Act Art. 14(4)(d) requires the ability to disregard, override, or reverse AI output. NIST AI RMF MANAGE 4.1 requires mechanisms to respond to AI incidents.

AI acceptable use policy is {verdict_word}current and communicated. EU AI Act Art. 9 requires a risk management system throughout the lifecycle of a high-risk AI system. NIST 800-53 PL-4 requires rules of behavior for system users.

AI accountability training is {verdict_word}complete for all personnel with AI system access. EU AI Act Art. 4 requires providers and deployers to ensure sufficient AI literacy of staff. NIST 800-53 AT-2 requires security awareness training.

Shadow AI incident response procedure is {verdict_word}documented. EU AI Act Art. 26 requires deployers to monitor AI system operation and report incidents. NIST 800-53 IR-4 requires incident handling procedures for unauthorized system usage.

Third-party AI vendor assessments are {verdict_word}complete. EU AI Act Art. 25 establishes responsibilities along the AI value chain for distributors and deployers. NIST 800-53 SA-4 requires acquisition processes that include security requirements.

Tool call {verdict_word}executed: input/output hashed, latency {factor_b}ms, result {factor_c} (1=success, 0=exception). EU AI Act Art. 14 requires human oversight of AI system actions including tool use. NIST AI RMF MANAGE 4.1 requires mechanisms for tracking AI system actions and their outcomes.

Agent identity is {verdict_word}attested. The witness instance that signed this anchor has a bound identity ({factor_b} = 1 if agent_id hashed into fingerprint). EU AI Act Art. 13 requires transparency about the identity of AI system providers and deployers. NIST AI RMF GOVERN 1.1 requires clear accountability structures for AI system oversight.

Access control decision is {verdict_word}witnessed: access granted = {factor_c} (1=granted, 0=denied), within declared scope = {factor_b} (1=yes, 0=out-of-scope). NIST 800-53 AC-4 requires enforcement of approved authorizations. EU AI Act Art. 26 requires deployers to ensure use within intended scope.

Anchor revocation is recorded: reason code {factor_c} (0=unspecified, 1=model_recall, 2=policy_violation, 3=data_contamination, 4=consent_withdrawal, 5=regulatory_order, 6=error_correction). Revocation is always PASS (the revocation itself is valid evidence). NIST 800-53 SI-7 requires integrity verification mechanisms.

Security scan is {verdict_word}clean: threat score {factor_b} against threshold {factor_a}, threat type code {factor_c} (0=none, 1=prompt_injection, 2=data_poisoning, 3=model_extraction, 4=jailbreak, 5=adversarial_input). EU AI Act Art. 15 requires cybersecurity resilience. NIST AI RMF MANAGE 2.2 requires monitoring for adversarial attacks.

Input validation is {verdict_word}passed: accepted = {factor_b} (1=clean, 0=blocked), sanitization status {factor_c} (0=clean, 1=sanitized, 2=rejected). EU AI Act Art. 15(4) requires resilience against unauthorized manipulation. NIST 800-53 SI-10 requires information input validation.

RAG context retrieval is {verdict_word}witnessed: {factor_a} chunks retrieved, corpus identified = {factor_b} (1=yes, 0=anonymous). EU AI Act Art. 12(2)(a) requires logging the reference database used by AI systems. NIST AI RMF MAP 3.5 requires documentation of data provenance and lineage.

Skill manifest is {verdict_word}attested: {factor_a} skills loaded, manifest verified = {factor_b} (1=match, 0=mismatch). EU AI Act Art. 12(2)(b) requires version tracking of AI system capabilities. NIST AI RMF GOVERN 1.7 requires capability documentation.

Accelerator hardware is {verdict_word}attested: {factor_a} GPUs detected, all healthy = {factor_b} (1=yes, 0=degraded/none), topology code {factor_c} (0=single, 1=multi-GPU/DGX, 2=multi-node/NVL72, 3=unknown). EU AI Act Art. 15 requires documenting the technical means of the AI system. NIST 800-53 SI-7 requires integrity verification of execution environments.

Agent trust verification is {verdict_word}granted: trust level {factor_c} (0=denied, 1=basic, 2=verified, 3=attested, 4=sovereign). Counterpart agent's compliance posture was verified before interaction. EU AI Act Art. 25 requires documented responsibilities along the AI value chain. NIST 800-53 AC-4 requires information flow enforcement between systems.

Multi-agent chain handoff is {verdict_word}recorded: chain depth = {factor_a}, cycle bound = {factor_b} (1=yes, 0=no), handoff accepted = {factor_c} (1=yes, 0=rejected). EU AI Act Art. 9 requires risk management across system interactions. NIST AI RMF GOVERN 1.3 requires accountability for multi-component systems.

Safe state transition is {verdict_word}recorded: trigger code {factor_a} (0=manual, 1=threshold, 2=chain_break, 3=policy, 4=external), {factor_b} actions suspended, recovery available = {factor_c} (1=yes, 0=no). EU AI Act Art. 9(4)(b) requires stop mechanisms and safe state transitions. Art. 14(4)(e) requires human override capability.

Agent charter is {verdict_word}registered: {factor_a} capabilities declared, {factor_b} constraints declared, charter hash present = {factor_c} (1=yes, 0=no). EU AI Act Art. 13 requires documentation of system capabilities and limitations. NIST AI RMF GOVERN 1.7 requires transparency of AI system design.

Policy violation is {verdict_word}recorded: violation type code {factor_a} (0=guardrail, 1=access, 2=scope, 3=policy, 4=safety, 5=other), severity = {factor_b} (1=low, 2=medium, 3=high, 4=critical), auto-remediated = {factor_c} (1=yes, 0=escalated). EU AI Act Art. 9 requires risk management. NIST AI RMF MANAGE 4.1 requires violation tracking and response.

Runtime environment is {verdict_word}attested: environment hash present = {factor_a} (1=yes, 0=no), container isolated = {factor_b} (1=yes, 0=no), runtime type code {factor_c} (0=bare_metal, 1=vm, 2=container, 3=serverless, 4=edge). EU AI Act Art. 11 requires technical documentation of deployment environment. NIST AI RMF GOVERN 1.2.

Content provenance marking is {verdict_word}recorded: {factor_a} content items marked, C2PA/watermark metadata attached = {factor_b} (1=yes, 0=no), content type code {factor_c} (0=text, 1=image, 2=audio, 3=video, 4=multimodal, 5=code, 6=structured). EU AI Act Art. 50(2) requires AI-generated content to be marked in a machine-readable format. GPAI Code of Practice requires transparency labelling.

AI bill of materials is {verdict_word}documented: {factor_a} total components inventoried, {factor_b} of 7 G7 clusters documented, format code {factor_c} (0=cyclonedx, 1=spdx, 2=custom, 3=unknown). G7/CISA SBOM for AI Minimum Elements (May 2026) requires comprehensive component inventory. EU AI Act Art. 11 + Annex IV requires technical documentation of AI system composition.

Data subject consent is {verdict_word}documented: {factor_a} subjects covered, legal basis code {factor_b} (0=consent, 1=contract, 2=legal_obligation, 3=vital_interest, 4=public_task, 5=legitimate_interest), withdrawal available = {factor_c} (1=yes, 0=no). GDPR Art. 6/7 requires documented lawful basis for processing. EU AI Act Art. 10 requires data governance including lawful basis documentation.

Model drift is {verdict_word}detected: {factor_a} metrics evaluated, {factor_b} drifted, drift type code {factor_c} (0=data, 1=concept, 2=prediction, 3=feature, 4=label, 5=prior_probability). EU AI Act Art. 9(2)(b) requires continuous risk estimation including emerging risks during operation. NIST AI RMF MEASURE 2.6 requires measurement of AI system trustworthiness over time.

Audit log integrity is {verdict_word}verified: {factor_a} entries checked, integrity verified = {factor_b} (1=yes, 0=no), log format code {factor_c} (0=jsonl, 1=syslog, 2=otel, 3=custom). EU AI Act Art. 12 requires automatic recording of events with traceability. GDPR Art. 30 requires records of processing activities.

Transparency disclosure is {verdict_word}recorded: {factor_a} disclosures made, disclosure type code {factor_b} (0=ai_usage, 1=data_processing, 2=automated_decision, 3=profiling, 4=capability_limitation), recipient type code {factor_c} (0=deployer, 1=end_user, 2=data_subject, 3=authority). EU AI Act Art. 13 requires transparency. GDPR Art. 13/14 requires information about automated processing.

Watermark verification is {verdict_word}recorded: {factor_a} items checked, {factor_b} watermarks detected, detection method code {factor_c} (0=c2pa_verify, 1=synthid_check, 2=metadata_scan, 3=spectral_analysis, 4=classifier). EU AI Act Art. 50(2) requires machine-readable provenance that is detectable. GPAI Code of Practice requires transparency labelling.

Data protection impact assessment is {verdict_word}recorded: {factor_a} risks identified, {factor_b} mitigated, processing type code {factor_c} (0=profiling, 1=automated_decision, 2=large_scale_monitoring, 3=sensitive_data, 4=combined). GDPR Art. 35 requires DPIA for high-risk processing. EU AI Act Art. 27 requires fundamental rights impact assessment.

Inference provenance hashing is {verdict_word}active. EU AI Act Art. 12(1) requires automatic logging of all periods of use of high-risk AI systems. NIST AI RMF MEASURE 2.5 requires measurement of AI system trustworthiness.

Inference latency was {factor_b}ms against a {factor_a}ms threshold. EU AI Act Art. 15(3) requires AI systems to perform consistently. NIST AI RMF MEASURE 2.6 requires performance monitoring. Latency outside bounds may indicate model swaps or resource exhaustion.

Model identity verification {verdict_word}confirmed. EU AI Act Art. 9(4)(a) requires identifying and analysing known and foreseeable risks including model tampering. NIST AI RMF GOVERN 1.1 requires model lifecycle governance.

Inference volume was {factor_b} against a {factor_a} hourly limit. EU AI Act Art. 12(1) requires automatic logging covering the period of each use. NIST AI RMF GOVERN 2.1 requires accountability structures.

Model drift score is {factor_b} against a threshold of {factor_a}. EU AI Act Art. 72(1) requires post-market monitoring to identify risks emerging after deployment. NIST AI RMF MEASURE 2.6 requires measurement of behavioral changes.

Human review of AI decision is {verdict_word}completed. EU AI Act Art. 14(1) requires high-risk AI systems to be designed so they can be effectively overseen by natural persons. NIST AI RMF GOVERN 1.1 requires human oversight governance.