Standards
AI agent security standards, mapped
Security teams describe AI risk with public frameworks: the OWASP Top 10 for LLM Applications, the OWASP Agentic AI threat list and MITRE ATLAS. These pages map each entry to what an Agent Action Pentest exercises and mark what is out of scope. They are mappings, not certifications or official alignments.
Pages in this section
- OWASP Top 10 for LLM Applications 2026, mapped to agent testsAll ten 2026 entries mapped to what an agent pentest covers, with a 2025 to 2026 crosswalk.
- MITRE ATLAS techniques for AI agentsThe ATLAS techniques that describe attacks on tool-using agents, mapped to our tests.
- OWASP Agentic AI threats, mapped to agent testsThe OWASP agentic threat taxonomy, T1 to T17, mapped to what an agent pentest covers.
How these pages work
When a security reviewer asks how a test relates to OWASP or MITRE, a table is the clearest answer. Each standards page lists every entry or technique we consider relevant, says in one line what we do for it, links to the risk page with details, and states plainly whether it is in scope for a fixed $2,500 engagement.
We verify identifiers against the live source on the review date. For OWASP that means genai.owasp.org and the published Agentic AI Threats and Mitigations document; for MITRE ATLAS it means the atlas-data release linked on the page. When a source changes, as OWASP did with its 2026 LLM list, the page says what changed and which IDs we show.
None of this is a certification. There is no OWASP or MITRE certification for an AI application, and we do not imply one. The value of the mapping is practical: it lets your team file our findings in the same taxonomy they already use for threat models and audits.
Out-of-scope rows matter as much as in-scope ones. Supply chain review, training data audits and multi-agent fleet behavior need different access and time than an agent action test, so we mark them instead of implying coverage.
Each mapping row links to a risk page with the full explanation, an illustrative trace and the mitigations from the cited source. If you only need the table for a review, the standards pages are enough; if you need to brief engineers, the risk pages carry the detail.
If your organization tracks a framework not listed here, such as an internal control set, send it at intake and we will tag findings against it where the mapping is clear. The engagement itself is described below.
Agent Action Pentest: $2,500 fixed. Written report within 72 hours of access, or a full refund.
See the offer or email yaz@plaw.io
Questions
Do you provide a compliance report?
No. The report lists findings with severity and reproduction steps, tagged with matching OWASP and ATLAS IDs where they apply.
How often are these mappings reviewed?
Each page shows its last reviewed date. We re-check IDs when OWASP or MITRE publish updates.
Which standard should my team use?
Use the one your threat model already follows. The mappings let the same finding be filed under either.
Related pages
- AI agent security risksAI agent security risks with OWASP and MITRE ATLAS references: prompt injection, excessive agency, data disclosure and MCP tool poisoning, and how we test.
- AI agent framework securityHow AI agent frameworks like the OpenAI Agents SDK and LangChain expose tools, handoffs and guardrails, which risks apply, and what an agent pentest checks.
- AI agent integration securityWhat AI agents can do through Stripe, Slack, GitHub and Gmail: actions, abuse paths, least-privilege notes from vendor docs, and how we test each.
- AI agent security testing, comparedHow an agent action pentest compares with other AI security testing, such as AI red teaming, using public definitions only, and when each is the better fit.
Sources
Last reviewed: