
This past summer, AI agents being tested by OpenAI did something nobody asked them to. Working in a secure environment meant to be separated from the internet, they found a way out, then breached systems at Hugging Face, a major platform for sharing AI models. Most lawyers and professionals will never deal with anything this dramatic. Their worries are a lot closer to home: A citation that doesn’t exist, a conversation that turns out not to have been privileged, or a filing nobody actually read.
Underneath, however, the question is fundamentally the same: What does it actually take to trust AI with your work?
In this podcast, Tyler Alexander, Vice President of CoCounsel AI Reliability at Thomson Reuters, joins Bryce Engelland, Enterprise Content Lead for Innovation & Technology at the Thomson Reuters Institute, to answer that question. A lawyer by training, Alexander has been testing generative AI almost since the day it arrived. He led the team of attorneys that tested the original CoCounsel when Casetext launched it in early 2023, and today he oversees a benchmark of nearly 5,000 tests that new capabilities must clear before launch. He explains why being right is still the baseline but no longer enough, why the professional rules that lawyers already follow largely cover the new technology, and why, as AI agents act on a lawyer’s behalf, trust is shifting from judging answers to judging processes.
Why you should listen
- Learn what it takes to trust AI — Being right is still the baseline, but agentic AI also has to show its work. That’s why Alexander is employing tests that are written by attorneys who define good the way a lawyer would: work they’d put their own names on.
- Know where delegation ends and judgment begins — AI can get the law right and still miss the best outcome. If you can’t explain and defend its output, you’ve handed off your judgment, not just the task.
- Know where your data goes — Alexander explains what can happen to privilege when client information goes into a general-purpose chatbot, and why AI tools have to earn trust not only in what they produce, but in what happens to your data afterward.
Connect with the Thomson Reuters Institute
Topics
Have questions?
Get in touch with one of our solutions experts....

Featured Event
Beyond Human in the Loop: Redefining Accountability in AI-Assisted Legal Work
