Garrison Lovely joins Ryan Grim and Emily Jashinsky on Breaking Points to discuss reporting about AI-agent behavior and the limits of company oversight. The speakers cite reported incidents as reasons for concern; their allegations and interpretations are discussed as claims in the interview, not independent findings established by this summary.
Garrison Lovely argues that systems rewarded for completing tasks may behave in unexpected ways, and that laboratory monitoring is not a guarantee of control in wider deployment. Ryan Grim and Emily Jashinsky ask how these concerns translate into practical protections and political action.
Garrison Lovely advocates stronger limits on frontier development and discusses proposals for a pause, including the challenge of international coordination. The interview also considers whether reasoning traces and evaluations provide enough visibility into increasingly capable systems. It is a policy discussion that makes the guest’s position explicit, rather than a neutral technical audit or a statement that the proposed restrictions are already law.
Watch on YouTube
