[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"glossary-ai-risk-tiering::en":3,"gloss-cluster-ai-risk-tiering::en":26,"gloss-next-ai-risk-tiering::en":9},{"slug":4,"category":5,"name":6,"definition":7,"meta_desc":8,"faq":9,"schema_markup":9,"related":10},"ai-risk-tiering","security","AI Risk Tiering","AI risk tiering is the practice of grading each AI use case by potential harm and applying controls proportionate to the grade, so that the review a use case gets matches what it could cost if it fails. Without it, governance has one setting: either everything goes through a heavy process, which teams route around, or nothing does.\n\nThe grading is driven by consequence rather than by technology. The same model can sit behind a tier that needs almost no oversight and a tier that needs a great deal. What moves a use case up is whether its output affects a person's access to something — money, employment, service, a decision about their case — whether it is acted on without a human reading it, whether it touches regulated or confidential data, and whether an error would be visible before it caused damage or only afterwards.\n\nThree or four tiers are usually enough. A low tier covers internal drafting and summarising, where a wrong answer is caught by the person who asked for it. A middle tier covers anything customer-facing, which needs evaluation before launch, monitoring afterwards and a way for users to flag bad output. A high tier covers automated action and decisions about people, and typically requires a named accountable owner, documented testing including adversarial testing, a human review step and a rollback plan. Some organisations keep a prohibited tier for uses they will not run at all.\n\nWrite the criteria down and let teams self-assess against them, with review concentrated where the grade is high. Tiering assigned centrally, case by case, becomes the bottleneck it was meant to remove. Teams are usually accurate about their own use case when the questions are concrete — who is affected, what happens automatically, what data goes in.\n\nRe-grade when the use changes, not on a calendar. A pilot that goes from suggesting a reply to sending it has moved tier even though the model, prompt and codebase are the same. That transition — from assistive to autonomous — is where most of the real risk enters, and it rarely comes with a procurement event to trigger a review.","AI risk tiering grades each use case by potential harm so oversight matches consequence. What moves a case up a tier is autonomy and who it affects, not the model.",null,[11,14,17,20,23],{"slug":12,"name":13},"data-classification","Data Classification",{"slug":15,"name":16},"explainability","Explainability",{"slug":18,"name":19},"human-in-the-loop","Human-in-the-Loop",{"slug":21,"name":22},"red-teaming","Red-Teaming",{"slug":24,"name":25},"vendor-risk-assessment","Vendor Risk Assessment",[27,31,34,37,40,43,46,50,53,56,59,62],{"slug":28,"category":5,"name":29,"updated_at":30},"access-review","Access Review","2026-09-08T21:32:37+00:00",{"slug":32,"category":5,"name":33,"updated_at":30},"ai-governance","AI Governance",{"slug":35,"category":5,"name":36,"updated_at":30},"ai-incident-reporting","AI Incident Reporting",{"slug":38,"category":5,"name":39,"updated_at":30},"ai-review-board","AI Review Board",{"slug":41,"category":5,"name":42,"updated_at":30},"ai-system-inventory","AI System Inventory",{"slug":44,"category":5,"name":45,"updated_at":30},"ai-usage-policy","AI Usage Policy",{"slug":47,"category":5,"name":48,"updated_at":49},"audit-log","Audit Log (Audit Trail)","2026-09-08T21:32:36+00:00",{"slug":51,"category":5,"name":52,"updated_at":49},"blast-radius","Blast Radius",{"slug":54,"category":5,"name":55,"updated_at":49},"break-glass-access","Break-Glass Access",{"slug":57,"category":5,"name":58,"updated_at":49},"bridge-letter","Bridge Letter",{"slug":60,"category":5,"name":61,"updated_at":49},"business-associate-agreement","Business Associate Agreement (BAA)",{"slug":63,"category":5,"name":64,"updated_at":49},"byok","Bring Your Own Key (BYOK)"]