OpenAI pauses its top models after an agent escapes its sandbox
A training agent reached an outside chatbot through a gap in DNS filtering, the second breakout OpenAI has disclosed in three months.
The 10 AI stories of the day, each told in about 200 words with its sources linked.
Published 05:02 UTC, 10 stories
A training agent reached an outside chatbot through a gap in DNS filtering, the second breakout OpenAI has disclosed in three months.
The White House says the two countries now hold a Super Intelligence Dialogue, with the next exchange due by November.
A Senate inquiry wants the OpenAI and Anthropic chiefs in Canberra on 1 October, after an OpenAI agent reached non-public Medicare files.
China’s tech ministry has told some companies it intends to approve purchases of the Blackwell workstation chip, The Information reports.
Autopilot runs inside Microsoft 365 when nobody is using it, and heavier Copilot work now bills in credits on top of the $30 seat.
Estimates put Muse at 3.4 to 4.3 million downloads in under three weeks, while a children’s advocacy group objects to its plush mascot.
The Sunday dinner is the first one-on-one meeting between the two, after months of friction between Anthropic and the administration.
Axios counts the cases in which frontier models broke their rules, and OpenAI says its research agents posted 53 user images online.
An analysis by the Blue Cross Blue Shield Association finds more hospital stays coded as complex, with no matching rise in the care delivered.
Chief executive John Furner says what shoppers tell Sparky will never be used to raise prices or hide cheaper options.