AAIF Reading Group: Measuring Agent Policy Compliance

MEETING

4:00 PM - 5:00 PM GMT

October 9, 2026

AAIF Reading Group: Measuring Agent Policy Compliance

How do we measure whether an AI agent is actually following its intended policies?

Join the Agentic AI Foundation community for an informal reading group exploring the paper “Measuring Agent Policy Compliance.”

Paper: https://arxiv.org/abs/2607.25398

We’ll read through the paper together, unpacking its approach to measuring policy compliance in AI agents, examining its key ideas and assumptions, and discussing the challenges involved in evaluating agent behavior.

This is not a formal presentation or lecture. It’s a community conversation: bring your questions, interpretations, disagreements, and ideas. We’ll leave plenty of room for open discussion and constructive debate.

We’ll explore:

How agent policy compliance can be measured

The paper’s methodology and key ideas

Challenges in evaluating whether agents follow their intended policies

Assumptions and limitations of the proposed approach

Questions and perspectives from the community

What this could mean for building and evaluating more reliable AI agents

If you’ve read the paper already, come ready to share your perspective. If you haven’t, you’re still welcome - we’ll work through it together.

Come curious. Bring questions. Let’s read the paper together and see where the discussion takes us.

Speakers

Arthur Coleman

CEO @ Online Matters

Roman Mkrtchian

Staff Backend Engineer @ Independent / Freelance

Apurva Misra

AI Consultant @ Sentick

Agenda

4:00 PM

4:05 PM

GMT

Opening / Closing

Welcome & framing

Introduce the paper, what we mean by agent policy compliance, and how we’ll structure the discussion

+ content:video.read_more_button

4:05 PM

4:25 PM

GMT

Presentation

Apurva Misra

Introduction, task, environment, harness & experiment, followed by discussion and questions around the methodology, assumptions, and evaluation setup.

+ content:video.read_more_button

4:25 PM

4:45 PM

GMT

Presentation

Roman Mkrtchian

Results, failure analysis, related research & solutions, followed by discussion around the findings, failure modes, and alternative approaches.

+ content:video.read_more_button

4:45 PM

5:00 PM

GMT

Fireside Chat

Open Community Discussion & Wrap-up

Bring the two sections together, discuss the broader implications for agent evaluation and policy compliance, and share remaining questions, disagreements, and takeaways.

+ content:video.read_more_button