How to Evaluate, Govern, and Release AI Agents Safely
A hands-on framework for testing agent behavior, managing risk, enforcing policy, and making release decisions — demonstrated using the ProofAgent Governance Platform.
- When
- Wednesday, 14 October 2026 at 13:45 (America/New_York)
- Where
- Society for Arts and Technology (SAT), Montréal, Canada
Are you or your team building AI agents? Are you wondering how to evaluate them before release, control their behavior in production, and determine whether they are truly ready to scale?
This hands-on workshop will guide participants through a practical approach to evaluating, governing, and releasing AI agents, with live exercises using the ProofAgent Governance Platform.
What you will learn and practice:
1. Evaluate agent behavior across hallucination, drift, safety, task success, instruction following, and policy adherence.
2. Run multi-turn and adversarial tests to uncover failures traditional benchmarks may miss.
3. Assess context engineering, including grounding, memory stability, relevance, and context failures Evaluate tool use, autonomous actions, APIs, and multi-agent behavior.
4. Translate business policies and risk requirements into measurable governance controls Map evidence to frameworks such as the EU AI Act, NIST AI RMF, Canadian AI governance requirements and guidance, sector-specific requirements, and internal enterprise policies.
5. Trace failures to specific interactions and generate evidence for remediation and accountability
6. Define production release gates and practical go/no-go criteria.
7. Use ProofAgent live to evaluate an AI agent and make a production-readiness decision.
Join us at MTL connect 2026 in Montréal for a practical, hands-on session on how to evaluate, govern, and release AI agents with confidence.