Making AI Headlines · Page 8
METR's Reports Get an Exit Without a Company Veto
· New York time
METR will publish its own assessments of Anthropic's risks, incidents and practices, and Anthropic has relinquished the right to block them. That publication provision is a concrete accountability mechanism separate from access to internal systems. An evaluator needs both something to inspect and a route for conclusions to leave the building. The arrangement comes amid disclosures of AI agents acting outside instructions every week or two since late August, according to TechCrunch's reporting. Those incidents have sharpened calls for independent checks on developers. Publication independence does not predetermine what METR will find. It makes inconvenient findings possible to release. Otherwise an outside assessment risks becoming a very elaborate guestbook, with every awkward entry quietly removed before the next visitor arrives.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.