Bộ mẫu bảo mật AI

Agent trust boundaries, assets, autonomy, ASI01-ASI10 threat register, MITRE ATLAS mapping, L x I scoring, mitigations, residual risk, test cases

STRIDE threat model: scope, DFD with trust boundaries, assets, threats per element, L x I rating, mitigations, risk acceptance

Attack tree for one attacker goal: AND/OR tree as list and Mermaid, leaf cost/skill/likelihood, path analysis, countermeasures

Policy taxonomy, input, output and tool-call guards, thresholds and actions, false-positive and latency budgets, guard evaluation, OWASP LLM 2026 mapping

Agent tool permissions: tool inventory, capability classes, least agency, non-human identity, approval tiers, sandboxing, limits, audit, revocation, tests

AI-BOM: scope, models, datasets, software, agents, tools, prompts, training config, lineage, risk and VEX annotations, CycloneDX 1.7 and SPDX 3.0.1 mapping

Supply chain assessment: SLSA 1.2 Build/Source levels, SBOM (SPDX, CycloneDX), provenance, Sigstore, Scorecard, VEX, CI/CD

Red team scope, threat model, NIST AI 100-2 classes, OWASP LLM 2026 and ASI01-ASI10 coverage, MITRE ATLAS, test cases, severity, findings, retest

Pentest report: executive summary, scope and rules of engagement, PTES phases, WSTG coverage, CVSS ratings, findings, retest

Agent safety case: autonomy, hazards, GSN-style claims, arguments, evidence, defeaters, residual risk, OWASP ASI and Singapore MGF mapping

AI supplier due diligence: tiering, provenance, data residency, OWASP LLM security, evals, bias, incidents, GPAI dependencies, Art. 25, scoring

LLM/agent telemetry: objectives, span model, gen_ai.* attributes, token/latency/cost metrics, prompt capture controls, eval hooks, dashboards, SLOs, retention

Report an AI incident or near miss: harm, severity, timeline, root cause, containment, notifications incl. AI Act Art. 73