AI Product Security Assessment
Assessment vocabulary for AI security evidence work.
AIPSA is the assessment vocabulary and benchmark domain model used inside the AI Security Scorecard, Field Guide, and Evidence-phase evidence work.
Credential levels
Four practitioner tiers.
70–79 score
Associate
Demonstrates foundational knowledge across AI Product Security domains.
Recommended for: Practitioners getting started
View certification →80–89 score
Practitioner
Proficient understanding of AI Product Security and practical application.
Recommended for: Security engineers, analysts, builders
View certification →90–94 score
Advanced
Advanced knowledge and ability to design, assess, and improve AI security controls.
Recommended for: Security leads, senior engineers, architects
View certification →95–100 score
Distinguished
Expert-level mastery across domains and leadership in AI Product Security.
Recommended for: Leaders, advisors, consultants, researchers
View certification →Assessment domains
14 domains across 6 groups.
Inventory & Architecture
What AI systems exist, how they're connected, who controls them, and where the trust boundaries are.
Explore evidence instruments →Adversarial Testing
Whether adversarial inputs, manipulation, and prompt injection are modeled, tested, and measured over time.
Explore evidence instruments →RAG & Data Authorization
Whether retrieval systems enforce proper authorization, and whether data exposure is controlled end to end.
Explore evidence instruments →Agentic Permissions
Whether agentic systems are constrained to the actions they need, with proper guardrails on tool use.
Explore evidence instruments →Detection & Incident Response
Whether AI systems produce actionable telemetry, and whether teams can detect and respond to AI-specific incidents.
Explore evidence instruments →Governance Evidence
Whether AI security is documented, auditable, and integrated into delivery — not just asserted.
Explore evidence instruments →AIPSA verification records confirm completion of a scoped assessment, lab path, or evidence review where issued. They do not certify that any product, organization, or system is free of vulnerabilities.