{ "notice": "Illustrative mapping from the AI Governance Engineer Body of Knowledge v0.5.0 (not a claim of conformity)", "version": "0.5.0", "license": "CC BY 4.0", "licenseUrl": "https://creativecommons.org/licenses/by/4.0/", "schemaVersion": 1, "schema": "https://aigovernanceengineer.com/api/v1/schemas/control.json", "self": "https://aigovernanceengineer.com/api/v1/controls/aige-ctl-deploy-012.json", "source": "https://aigovernanceengineer.com/controls/deployment-and-monitoring#aige-ctl-deploy-012", "citation": { "title": "AI Governance Engineering: The Thesis & Body of Knowledge", "authors": [ "Jorge García Aibar" ], "parentDoi": "https://doi.org/10.5281/zenodo.22956197", "conceptDoi": "https://doi.org/10.5281/zenodo.22857084" }, "control": { "id": "AIGE-CTL-DEPLOY-012", "profile": "deployment-and-monitoring", "url": "https://aigovernanceengineer.com/controls/deployment-and-monitoring#aige-ctl-deploy-012", "json": "https://aigovernanceengineer.com/api/v1/controls/aige-ctl-deploy-012.json", "title": "Deactivation triggers, degraded modes and suspension", "version": "0.1", "status": "draft", "reviewerStatus": "open", "depth": "derived", "objective": "A runbook kept with the registry entry names the threshold and legal triggers for stopping the system, each with the role that decides and the time allowed; every path first freezes the logs, applies a legal hold and snapshots the pinned versions; degraded modes short of shutdown, and a suspension path for systems the deployer does not own, are built as tested toggles and drilled at least yearly.", "failureModes": [ "A trigger fires and nobody knows who may decide, so the system keeps serving while a meeting runs.", "Logs are overwritten before the evidence is frozen.", "The only available action is to turn everything off, everywhere, which is often worse than the fault.", "A classifier embedded in a vendor product has no switch, so a deployer with reason to consider that it presents a risk cannot suspend its use.", "A degraded mode or the suspension path fails the first time it is used, because it was never drilled." ], "scope": "Every deployed AI system that a breached floor, an incident or a legal duty could force to stop, including systems embedded in a supplier's product. For agents, the kill switch and circuit breaker controls of the Agent Runtime profile (AIGE-CTL-AGENT-011 and AIGE-CTL-AGENT-010) are the instant form.", "enforcementPoints": [ "runtime", "periodic" ], "verification": [], "evidence": [ { "artefact": "Runbook with triggers, deciding roles and times allowed; toggle change log; drill records with the time to decide, to the degraded mode and to off", "schemaId": null, "schema": null, "layer": 4 } ], "failureResponse": { "effect": "deny", "text": "On a trigger the decision owner switches the system to a degraded mode or off within the time allowed, after the evidence is frozen; under Art. 26(5) the deployer suspends use and informs the provider or distributor and the authority." }, "layer": 4, "secondaryLayers": [], "patterns": [ { "slug": "deactivation-localisation-retirement-runbook", "title": "Deactivation, Localisation & Retirement Runbook", "url": "https://aigovernanceengineer.com/patterns/deactivation-localisation-retirement-runbook" }, { "slug": "incident-pipeline", "title": "Incident Pipeline", "url": "https://aigovernanceengineer.com/patterns/incident-pipeline" } ], "seeds": [], "derivedFrom": [ { "kind": "pattern", "ref": "deactivation-localisation-retirement-runbook", "url": "https://aigovernanceengineer.com/patterns/deactivation-localisation-retirement-runbook" }, { "kind": "pattern", "ref": "incident-pipeline", "url": "https://aigovernanceengineer.com/patterns/incident-pipeline" }, { "kind": "chapter", "ref": "governing-deployment", "url": "https://aigovernanceengineer.com/bok/governing-deployment" }, { "kind": "chapter", "ref": "incidents", "url": "https://aigovernanceengineer.com/bok/incidents" } ], "mappings": { "obligations": [ { "id": "AIGE-OBL-EUAIA-ART26-5", "name": "EU AI Act Art. 26(5) deployer monitoring, suspension and informing the provider", "url": "https://aigovernanceengineer.com/obligations/aige-obl-euaia-art26-5" }, { "id": "AIGE-OBL-EUAIA-ART20", "name": "EU AI Act Art. 20 corrective actions and duty of information", "url": "https://aigovernanceengineer.com/obligations/aige-obl-euaia-art20" }, { "id": "AIGE-OBL-EUAIA-ART5", "name": "EU AI Act Art. 5 prohibited practices (incl. new NCII and CSAM bans)", "url": "https://aigovernanceengineer.com/obligations/aige-obl-euaia-art5" } ], "iso42001": [ { "id": "A.6.2.6", "title": "AI system operation and monitoring" } ], "nistAiRmf": [ { "id": "MANAGE 2.4", "title": "Mechanisms to supersede, disengage or deactivate AI systems" } ], "owasp": [], "atlas": [], "aiuc1": [], "csaAicm": [], "other": [] }, "references": [ { "n": 44, "title": "Pattern: Deactivation, Localisation & Retirement Runbook", "text": "Pattern: Deactivation, Localisation & Retirement Runbook (AI Governance Engineering Body of Knowledge v0.5.0, pattern catalogue (chapter 05)). AI Governance Engineer (Jorge García Aibar). 2026-09.", "url": "https://aigovernanceengineer.com/patterns/deactivation-localisation-retirement-runbook", "verified": "primary" }, { "n": 45, "title": "Governing deployment and use", "text": "Governing deployment and use (AI Governance Engineering Body of Knowledge v0.5.0, chapter 15, section \"A deactivation policy someone can execute\"). AI Governance Engineer (Jorge García Aibar). 2026-09.", "url": "https://aigovernanceengineer.com/bok/governing-deployment#a-deactivation-policy-someone-can-execute", "verified": "primary" }, { "n": 46, "title": "Governing deployment and use", "text": "Governing deployment and use (AI Governance Engineering Body of Knowledge v0.5.0, chapter 15, section \"Graduated degradation\"). AI Governance Engineer (Jorge García Aibar). 2026-09.", "url": "https://aigovernanceengineer.com/bok/governing-deployment#graduated-degradation", "verified": "primary" }, { "n": 9, "title": "Incidents, issues and root causes", "text": "Incidents, issues and root causes (AI Governance Engineering Body of Knowledge v0.5.0, chapter 17, section \"Deployer duties: inform the provider, suspend use\"). AI Governance Engineer (Jorge García Aibar). 2026-09.", "url": "https://aigovernanceengineer.com/bok/incidents#deployer-duties-inform-the-provider-suspend-use", "verified": "primary" }, { "n": 24, "title": "\"Feature Toggles (aka Feature Flags)\"", "text": "\"Feature Toggles (aka Feature Flags)\" (release, experiment, ops and permissioning toggles; ops kill switches for graceful degradation). Pete Hodgson, martinfowler.com. 2017-10-09.", "url": "https://martinfowler.com/articles/feature-toggles.html", "verified": "primary" }, { "n": 4, "title": "Regulation (EU) 2024/1689 (AI Act), consolidated text of 2026-07-27, Art. 26", "text": "Regulation (EU) 2024/1689 (AI Act), consolidated text of 2026-07-27, Art. 26 (deployer obligations: 26(1) use per the instructions; 26(2) oversight by competent persons with authority; 26(5) monitor, suspend and inform, serious incidents to the provider first; 26(6) logs kept at least six months). Publications Office of the EU (EUR-Lex). 2026-07-27.", "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/2026-07-27/eng#art_26", "verified": "primary" }, { "n": 47, "title": "Regulation (EU) 2024/1689 (AI Act), consolidated text of 2026-07-27, Art. 20", "text": "Regulation (EU) 2024/1689 (AI Act), consolidated text of 2026-07-27, Art. 20 (providers take corrective action: bring into conformity, withdraw, disable or recall; inform distributors and deployers). Publications Office of the EU (EUR-Lex). 2026-07-27.", "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/2026-07-27/eng#art_20", "verified": "primary" }, { "n": 48, "title": "Regulation (EU) 2024/1689 (AI Act), consolidated text of 2026-07-27, Art. 5", "text": "Regulation (EU) 2024/1689 (AI Act), consolidated text of 2026-07-27, Art. 5 (prohibited practices, binding on deployers as well as providers). Publications Office of the EU (EUR-Lex). 2026-07-27.", "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/2026-07-27/eng#art_5", "verified": "primary" }, { "n": 5, "title": "ISO/IEC 42001:2023, AI management systems, Annex A", "text": "ISO/IEC 42001:2023, AI management systems, Annex A (reference control objectives and controls A.2 to A.10, cited by id and short title). ISO/IEC. 2023.", "url": "https://www.iso.org/standard/81230.html", "verified": "secondary" }, { "n": 6, "title": "Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1", "text": "Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (subcategories cited by id: GOVERN 1.6, 1.7, 2.2, 6.1; MAP 1.1, 3.5; MEASURE 2.3, 2.4, 2.11, 3.1; MANAGE 1.1, 2.4, 3.1, 4.1, 4.3). NIST. 2023-01-26.", "url": "https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf", "verified": "primary" } ], "implementationNotes": [ "Build the intermediate modes in advance: advice-only, raised thresholds with abstention to a person, grounded-only answers, scoped off for one group, language, region or function, back to the pilot cohort, and off with the fallback process.", "Suspension is a kill switch for a system you do not own: you cannot revoke a vendor's weights, but you can stop sending it traffic; test that you can, and how long it takes.", "Keep the jurisdiction as a policy input, with flags by region, so one market can be switched off without touching the others." ], "openQuestions": [ "Verification procedure to be specified: the source material states what the control produces, not how a third party checks it; requires technical review.", "Where the model sits inside a supplier's product the switch depends on the contract; which terms give the deployer a switch is not settled here." ], "observation": null, "observationSchema": "https://aigovernanceengineer.com/schemas/control-observation.v1.json", "examples": [] } }