Making AI Headlines · Page 5
Six METR Researchers Get Past the Demonstration Room
· New York time
Anthropic says it is giving six researchers from METR employee-level access to assess its models, safety practices and risk management independently. METR is a nonprofit that tests frontier AI models. The arrangement gives outside evaluators a closer view of internal systems than a polished demonstration can provide. That access addresses an information gap between a developer and people asked to assess its risks. Models, operating practices and risk decisions are connected subjects, so examining only the public-facing product can leave important questions untouched. The agreement creates an opportunity for scrutiny, not a favorable result in advance. Inspecting a restaurant's kitchen is more informative than admiring its menu, even when the menu has been trained to explain how clean the kitchen probably is.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.