VIII. Help and legal

Frequently asked questions

What is ATR?: An open detection-rule standard for AI agent threats: a specification, a versioned set of YAML rules, and a test suite any engine can use to show it conforms.
Does it cost anything?: No. The rules and the specification are MIT licensed, so anyone may use, implement and modify them. Organization membership is not open yet. It never buys the outcome of a rule.
Who maintains it?: The maintainers, for now. This is a transition period while the Technical Steering Committee is being seated. See governance and the charter.
Does ATR certify products or organizations?: No. ATR does not audit or certify organizations or their deployments. Engines self-declare conformance against the public test suite. That is a self-declared L1 conformance, not a certification issued by ATR, and it has nothing to do with membership. Auditors can verify a deployment themselves with the auditor kit.
What is a lane?: A lane decides which rule maturities may fire. Enforce runs maturity stable only, and it is the lane to run with blocking on. Alert runs stable and test. Hunt runs every effective rule and is the default. Blocking is off by default in the engine and the Claude Code hook; the Mastra, OpenShell and NemoClaw adapters block high and critical matches by default.
Why are there three rule counts?: There are 844 rule files on main. 836 are effective, meaning at least one lane loads them. 8 are inert: their status is draft or deprecated, and no lane loads them. Any one number alone overstates or understates, so every figure on this site says what it counts.
Which version should I cite?: Cite the specification version for the specification, and the npm version for the software. They are labelled apart; formats are on the citation page.
Is ATR affiliated with OWASP or MITRE?: No. The crosswalks are ATR's own documents; the framework owners have not reviewed or endorsed them.
Does ATR send my data anywhere?: The reference engine evaluates rules inside your own process. Reporting to an outside service is off by default and has no built-in endpoint. Turning it on takes an explicit flag and a receiver you choose.
How do I report a security vulnerability?: Privately: through GitHub private vulnerability reporting, or by email to [email protected]. Do not open a public issue. The process is in SECURITY.md.
How do I report a false positive or a bypass?: With the GitHub false-positive form or the evasion form. Accepted reports are credited in the release notes.
How do I join?: Joining is paused until the privacy notice is final. Once it opens, joining is free, and the directory lists only people who agreed to be listed.

Page generated 2026-11-10Something wrong or missing? Open an issue