{ "aibom_controls": [ { "action": "Define included component classes, environments and discovery methods; enumerate inaccessible or unsupported components.", "change_trigger": "The system boundary or discovery method changes.", "control_id": "AIBOM-SCOPE", "evidence_type": [ "Declared scope and exclusion list", "Discovery coverage check" ], "name": "Inventory scope and blind spots", "owner_role": "AI platform owner", "risk": "An inventory is treated as complete although the scanner cannot observe some components.", "rollback": "Pause promotion for unresolved critical inventory gaps; restore the last understood boundary without claiming scanner completeness.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Compare one known component of each in-scope class against the inventory and record omissions." }, { "action": "Record stable component identifiers, release versions and available content digests; mark unavailable identities explicitly.", "change_trigger": "A model, package, prompt, data snapshot or build changes.", "control_id": "AIBOM-IDENTITY", "evidence_type": [ "Version and digest manifest", "Inventory diff" ], "name": "Immutable component identity", "owner_role": "Release engineering owner", "risk": "A mutable label hides a component change.", "rollback": "Block the unidentified component and restore a known compatible identity after verifying that it remains supported and authorized.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Change a fixture version or digest and confirm the inventory diff exposes it." }, { "action": "Record origin, producer, acquisition path and referenced provenance separately from verification status.", "change_trigger": "Origin, producer, acquisition path or provenance policy changes.", "control_id": "AIBOM-PROVENANCE", "evidence_type": [ "Origin references", "Provenance verification record" ], "name": "Traceable provenance", "owner_role": "Supply-chain security owner", "risk": "A component cannot be linked to its asserted origin.", "rollback": "Quarantine the component if required provenance cannot be verified; retain the evidence used for the decision.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Trace a selected component to the retained origin record; mark unverifiable assertions as unknown." }, { "action": "Record the applicable license or terms reference, proposed use, open questions and the human decision owner.", "change_trigger": "License, terms, intended use, data access or distribution changes.", "control_id": "AIBOM-RIGHTS", "evidence_type": [ "Version-specific terms reference", "Named rights and access decision" ], "name": "Permitted use and data access", "owner_role": "Rights reviewer with the data owner", "risk": "A release depends on a component or dataset whose allowed use is unresolved.", "rollback": "Block the affected use or distribution and select a reviewed permitted substitute; preserve the rights decision history.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Inspect the exact version and use decision; do not treat a scanner license label as approval." }, { "action": "When attestations are required, verify the subject digest, signature identity and expected producer policy before accepting the claim.", "change_trigger": "Subject digest, signer trust, producer policy or verification tooling changes.", "control_id": "AIBOM-ATTESTATION", "evidence_type": [ "Attestation and subject digest", "Trust-policy version", "Positive and negative verification records" ], "name": "Attestation subject and trust verification", "owner_role": "Supply-chain security owner", "risk": "A valid-looking attestation is accepted for the wrong artifact or producer.", "rollback": "Reject the untrusted attestation and block the component; re-admit only after verification against a reviewed policy.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Reject fixtures with a changed subject digest, untrusted signer or unexpected producer; retain the verification outcome." }, { "action": "Compare the candidate inventory with the approved baseline and route material changes to named component owners.", "change_trigger": "Any in-scope inventory item or its permissions change.", "control_id": "AIBOM-DIFF", "evidence_type": [ "Baseline and candidate inventories", "Material diff and decision log" ], "name": "Material inventory change review", "owner_role": "Release engineering owner", "risk": "A dependency change reaches production without an owned review.", "rollback": "Hold promotion and restore a compatible reviewed inventory after checking migrations and external effects.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Introduce a new component and a permission change in a fixture; confirm both appear in the review record." }, { "action": "Bind the inventory to an enforceable admission decision and a current revocation process.", "change_trigger": "Admission policy, trust status, exposure or revocation information changes.", "control_id": "AIBOM-ADMISSION", "evidence_type": [ "Admission policy and decisions", "Revocation exercise" ], "name": "Dependency admission and revocation", "owner_role": "Platform security owner", "risk": "A revoked or disallowed dependency remains available at runtime.", "rollback": "Revoke the affected capability and use a reviewed supported fallback; deployment rollback must not reintroduce revoked components.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Attempt admission of a permitted and a revoked fixture; confirm both decisions are attributable." }, { "action": "Assign a responsible maintainer and accountable release owner; define review triggers and retain change history.", "change_trigger": "Team ownership, release process, component lifecycle or evidence location changes.", "control_id": "AIBOM-OWNERSHIP", "evidence_type": [ "Owner assignment", "Review schedule and trigger record", "Change history" ], "name": "Inventory ownership and maintenance", "owner_role": "System owner", "risk": "Inventory evidence becomes stale without an accountable maintainer.", "rollback": "Pause unowned changes and assign an accountable replacement before reopening promotion.", "system_boundary": "The selected release and its model, prompt, data, tool and runtime dependency inventory.", "verification": "Simulate an ownership transfer and verify that review and escalation routes remain reachable." } ], "architecture": { "criteria": [ { "allowedAnswers": [ "unknown", "yes", "no" ], "core": true, "id": "evaluation_ready", "question": "Is there a representative evaluation set and an explicit acceptance decision?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": true, "id": "fixed_steps", "question": "Can the work be described as a stable sequence of steps?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": true, "id": "external_knowledge", "question": "Does the task need a controlled external knowledge source?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": true, "id": "tool_actions", "question": "Must the system perform external tool actions?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": true, "id": "independent_subtasks", "question": "Can subtasks be isolated with explicit input and output contracts?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": true, "id": "parallel_justified", "question": "Does a measured baseline justify the added coordination cost?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": false, "id": "approval_boundary", "question": "For tool actions, are authorization and human approval boundaries explicit?" }, { "allowedAnswers": [ "unknown", "yes", "no" ], "core": false, "id": "recovery_tested", "question": "For tool actions, has failure recovery been exercised?" } ], "limitations": [ "Original engineering proposal, not an empirical benchmark or a released canonical source.", "Patterns may be combined after evaluation; the first match is a starting point, not proof of fitness.", "The 2024 pattern reference is conceptual; current framework features must be verified separately." ], "method": "Ascending priority, first matching rule; all AND any AND none; unspecified answers are unknown.", "references": [ { "checked_at": "2026-09-17", "scope": "Conceptual distinction between fixed workflows and dynamic agents, and common composition patterns; not validation of these proposed rules.", "url": "https://www.anthropic.com/engineering/building-effective-agents" } ], "rules": [ { "id": "ARCH-NO-EVALUATION", "next_steps": [ "Define representative cases, failure criteria and an accountable acceptance owner before selecting an autonomous path." ], "outcome": "EVALUATION_REQUIRED", "priority": 10, "rejected_options": [ "AUTONOMOUS_RELEASE" ], "when": { "all": [ { "criterionId": "evaluation_ready", "operator": "equals", "value": "no" } ], "any": [], "none": [] } }, { "id": "ARCH-UNKNOWN-CORE", "next_steps": [ "Resolve the unknown core questions; unknown is not equivalent to no." ], "outcome": "INSUFFICIENT_EVIDENCE", "priority": 20, "rejected_options": [], "when": { "all": [], "any": [ { "criterionId": "evaluation_ready", "operator": "equals", "value": "unknown" }, { "criterionId": "fixed_steps", "operator": "equals", "value": "unknown" }, { "criterionId": "external_knowledge", "operator": "equals", "value": "unknown" }, { "criterionId": "tool_actions", "operator": "equals", "value": "unknown" }, { "criterionId": "independent_subtasks", "operator": "equals", "value": "unknown" }, { "criterionId": "parallel_justified", "operator": "equals", "value": "unknown" } ], "none": [] } }, { "id": "ARCH-ACTION-BOUNDARY", "next_steps": [ "Define scoped authorization, approval, containment and recovery for external effects." ], "outcome": "ACTION_BOUNDARY_REQUIRED", "priority": 30, "rejected_options": [ "AUTONOMOUS_TOOL_ACTIONS" ], "when": { "all": [ { "criterionId": "tool_actions", "operator": "equals", "value": "yes" } ], "any": [ { "criterionId": "approval_boundary", "operator": "not_equals", "value": "yes" }, { "criterionId": "recovery_tested", "operator": "not_equals", "value": "yes" } ], "none": [] } }, { "id": "ARCH-FIXED-WORKFLOW", "next_steps": [ "Implement explicit step transitions and validation gates; attach retrieval or approved tools only where the step requires them." ], "outcome": "DETERMINISTIC_WORKFLOW", "priority": 40, "rejected_options": [], "when": { "all": [ { "criterionId": "fixed_steps", "operator": "equals", "value": "yes" } ], "any": [], "none": [] } }, { "id": "ARCH-RETRIEVAL", "next_steps": [ "Validate source access, retrieval coverage and answer grounding against the evaluation set." ], "outcome": "RETRIEVAL_PIPELINE", "priority": 50, "rejected_options": [], "when": { "all": [ { "criterionId": "tool_actions", "operator": "equals", "value": "no" }, { "criterionId": "external_knowledge", "operator": "equals", "value": "yes" } ], "any": [], "none": [] } }, { "id": "ARCH-ORCHESTRATION", "next_steps": [ "Define worker boundaries, a coordinator merge contract, cancellation and duplicate-effect handling before an evaluation." ], "outcome": "ORCHESTRATOR_WORKERS", "priority": 60, "rejected_options": [], "when": { "all": [ { "criterionId": "independent_subtasks", "operator": "equals", "value": "yes" }, { "criterionId": "parallel_justified", "operator": "equals", "value": "yes" } ], "any": [], "none": [] } }, { "id": "ARCH-BOUNDED-AGENT", "next_steps": [ "Start with one agent and scoped tools, iteration limits and a tested stop path; compare it with the simplest working baseline." ], "outcome": "BOUNDED_TOOL_AGENT", "priority": 70, "rejected_options": [], "when": { "all": [ { "criterionId": "tool_actions", "operator": "equals", "value": "yes" } ], "any": [], "none": [] } }, { "id": "ARCH-BASELINE", "next_steps": [ "Evaluate a single-call baseline before adding dynamic coordination." ], "outcome": "SINGLE_CALL_BASELINE", "priority": 80, "rejected_options": [], "when": { "all": [ { "criterionId": "evaluation_ready", "operator": "equals", "value": "yes" } ], "any": [], "none": [] } } ], "version": "0.1.0-draft.1" }, "authoring": "AI-assisted preparation, 2026-09-17. No human signature or experimental outcome is asserted.", "governance": { "accountability": { "risk": "Release or incident decisions have no accountable person.", "rollback": "Pause decisions that require the missing authority and route them to the documented accountable replacement.", "trigger": "People, delegation, escalation or release authority changes." }, "data": { "risk": "Data is used outside its permitted scope or retained after revocation.", "rollback": "Remove access to the affected data and rebuild a permitted snapshot; never restore revoked data to improve quality metrics.", "trigger": "Sources, permissions, retention, population coverage or correction requests change." }, "incident": { "risk": "Containment or notification review is delayed by unclear responsibility.", "rollback": "Invoke the containment authority and preserve the timeline; route notification decisions to the named qualified reviewer.", "trigger": "Severity criteria, incident contacts, workflows or applicable notification requirements change." }, "inventory": { "risk": "Unregistered components invalidate a release decision.", "rollback": "Pause promotion until the register matches the actual release; restore only a documented compatible bundle.", "trigger": "A model, prompt, dataset, tool, purpose or environment changes." }, "literacy": { "risk": "Operators cannot recognize or contain a known failure.", "rollback": "Restrict affected operations to a trained supervisor until the role-specific exercise is completed.", "trigger": "A role, workflow, limitation or escalation procedure changes." }, "monitoring": { "risk": "A measured breach receives no owned or tested response.", "rollback": "Contain the affected capability and rehearse the compatible recovery path; investigate the breach before expanding traffic.", "trigger": "Signals, thresholds, alert routing, baseline or recovery targets change." }, "oversight": { "risk": "A human cannot understand, override or stop the workflow.", "rollback": "Disable autonomous action and route affected cases through an accountable human until override and stop exercises pass.", "trigger": "The interface, action authority, notices or operator workflow changes." }, "records": { "risk": "A material decision cannot be reconstructed within the retention policy.", "rollback": "Pause affected high-impact decisions while restoring the permitted trace path; preserve existing records without expanding sensitive-data retention by default.", "trigger": "Trace schema, approval linkage, storage, access or retention policy changes." }, "risk": { "risk": "A known failure is accepted without a current owner or authority.", "rollback": "Pause the exposed use case and reopen its risk decision; preserve objections and obtain an explicit acceptance or mitigation decision.", "trigger": "A new failure scenario, incident, mitigation change or residual-risk decision appears." }, "scope": { "risk": "A use or affected population falls outside the reviewed scope.", "rollback": "Suspend the expanded use and return to the previously reviewed scope while the assigned reviewer resolves open questions.", "trigger": "Intended use, geography, affected population or decision impact changes." }, "supplier": { "risk": "A supplier change breaks the operating or evidence boundary.", "rollback": "Disable the affected integration or move to a reviewed supported fallback; do not restore a retired or revoked dependency.", "trigger": "Supplier terms, supported versions, access, hosting or component ownership changes." }, "validation": { "risk": "A release is promoted without evidence for its intended use.", "rollback": "Stop promotion and rerun the same relevant evaluation against a compatible known-good candidate before routing traffic.", "trigger": "The release, rubric, evaluation set, threat model or acceptance threshold changes." } }, "references": [ { "checked_at": "2026-09-17", "scope": "Attestation verification concepts for AIBOM-ATTESTATION; no SLSA level or certification claim.", "url": "https://slsa.dev/spec/v1.1/verifying-artifacts" } ], "security_layers": { "identity-access": { "risk": "An unintended principal obtains another identity or scope.", "rollback": "Revoke affected credentials, deny the affected capability and restore a reviewed compatible authorization policy; verify tenant isolation before reopening access.", "trigger": "identity, audience, token lifetime, tenant binding or authorization policy changes." }, "input-instruction": { "risk": "Untrusted content crosses an instruction or data boundary.", "rollback": "Quarantine the affected input path and restore a compatible reviewed prompt/parser boundary; recheck legitimate and adversarial inputs before routing traffic.", "trigger": "prompt hierarchy, content source, parser, output schema or isolation boundary changes." }, "memory-integrity": { "risk": "Unauthorized or stale memory affects a later decision.", "rollback": "Quarantine affected entries and restore an authorized compatible memory snapshot; preserve the audit trail and do not restore deleted or revoked data.", "trigger": "memory namespace, writer policy, retention, provenance or retrieval behavior changes." }, "monitoring-oversight": { "risk": "An unsafe action continues without a usable audit, approval or containment path.", "rollback": "Stop the affected action route, preserve available evidence and use a read-only or human-mediated fallback until the audit and containment path is exercised.", "trigger": "logging, limits, approval scope, escalation routes or containment mechanisms change." }, "supply-chain": { "risk": "An unapproved component or capability enters the runtime.", "rollback": "Block the suspect component and restore a reviewed compatible dependency set; verify signatures, access scope and current revocation status before reuse.", "trigger": "component identity, digest, provenance, dependency graph, owner or admission policy changes." }, "tool-execution": { "risk": "A tool action exceeds its authorized scope or creates duplicate external effects.", "rollback": "Disable the affected write or execution capability. Restore a compatible restricted route after replay checks; separately assess compensation for completed external effects.", "trigger": "tool arguments, permissions, sandbox, network destinations or retry semantics change." } }, "status": "ENGINEERING_PROPOSALS_NOT_REVIEWED", "upstream_adoption": "PENDING; these are proposed additions, separate from pinned source files." }