Skip to content

BRK312 Breakout Wed 18 Nov 16:15 PST

The trust foundation for the next era of intelligence

Sarah Bird, Mohammad Abuomar, Bogdan Demeny, Andrew Welch

Summary

From the recording

Agent Controls was announced in public preview, with no date stated. It builds on the open-source agent control specification, applying deterministic rules, safety classifiers and LLM judges to runtimes compliant with agent hooks. In a bank demo with two manual examples, success rose from 73% (transfer approval) and 80% (hardship) to 100% for both. Microsoft said internal measurement showed frontier coding models completing tasks about 90% of the time but, in some cases, following constraints only 40% of the time. Centru, with Assert inside, was said to arrive in the December release. Captions are auto-generated, and no pricing was given.

Session description

Microsoft's catalogue description

AI is entering a new era where agents reason, collaborate, and take action across business systems. As intelligence becomes more autonomous, trust becomes the critical infrastructure that enables innovation to scale. Join Sarah and Mohammad to explore Microsoft's vision for the next generation of Trustworthy AI: from policy and oversight to security, transparency, and accountability. Learn how organizations can confidently unlock the promise of agentic AI while staying in control.

← All sessions