Frontier AI Models Are Losing Monitorability | Reilly Haskins
METR technical staff member Reilly Haskins discusses their new research note on live, per-action monitoring systems designed to catch hazardous actions during AI evaluations. He outlines the technical limits of frontier model monitoring, how fast-advancing capabilities degrade transparency, and how organizations can enforce automated AI safeguards. Check out our sponsor Lovable: https://lovable.dev. Turn ideas into software people love.
--https://x.com/ReillyHaskins02--
MTS is a live news and interview show covering technology, business, politics, and culture as it happens.
Follow MTS:
X |…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.