Hands-on workshops, conference talks, meetups and private training on AI agent governance: the ProofAgent Index (PAI), adversarial evaluation with the open-source ProofAgent Harness, and turning a compliance obligation into a test that either passes or fails. Delivered on site or remote, for engineering, platform and risk teams.
Every session is built on the four axes of the ProofAgent Index — evaluation, context, compliance and governance — published on arXiv (arXiv:2607.27677) as "Stop Shipping AI Agents on Faith: Capability Is Not Production Readiness". Attendees learn to score an agent's production readiness rather than its capability: a strong axis cannot rescue a weak one, all four axes are required or the verdict is withheld, and anything dangerous caps the score with the report naming what capped it.
Nothing taught is proprietary. Sessions run on the Apache 2.0 ProofAgent Harness (published to PyPI as proofagent-harness), with its 183 adversarial traps across 11 attack families and its screening against a catalog of 25 regulatory frameworks. Attendees wire up their own agent, run multi-turn adversarial evaluation against it, read the evidence-linked findings, and leave able to repeat the whole exercise on their own machine.
Any session can be delivered privately and rebuilt around your own agents, your own regulatory scope and your own release process. Tell us the audience, the timeframe and the topics and you get a concrete agenda back rather than a brochure.