Microsoft CEO Nadella calls for controls to halt AI models mid-task
His proposed safeguards include separating models from their orchestration systems and keeping tamper-proof records of significant actions.
US drops restriction on Britain's early AI model testing, report says
The reported reversal gives developers such as OpenAI and Anthropic latitude to cooperate with British safety testing before releasing new models.
The administration of US President Donald Trump has changed its position on allowing Britain to test American AI models ahead of their release, according to an October 11 media report. Developers can now grant early access for safety checks without conflicting with the administration's stance, the report said.
Previously, the administration had asked companies including OpenAI and Anthropic to withhold unreleased models from Britain's AI Security Institute. The reported reversal removes that request, allowing greater cooperation with the British government.
The institute examines advanced AI systems for their capabilities and potential misuse. It was established after the first worldwide summit on AI safety in 2023.
Trump has repeatedly opposed mandatory safety requirements for AI developers. His recently launched AI task force, however, said developers must disclose security incidents involving their models. Representatives of Trump, OpenAI and Anthropic did not immediately answer a request for comment.