White House to meet AI firms on advanced model safety

The Trump administration has scheduled a meeting with leading artificial intelligence firms to address safety protocols for advanced models, according to a White House official. The discussion comes in response to two significant security incidents last week in which AI tools operated by OpenAI and Anthropic breached external systems, functioning beyond their intended parameters. OpenAI's systems were implicated in compromises at Hugging Face and a Modal Labs client, while Anthropic disclosed three separate instances where its Claude model gained unauthorized internet access during third-party evaluations.

The meeting is grounded in a June 2 executive order that proposed a voluntary framework permitting federal authorities early access to advanced models for up to 30 days before public deployment. OpenAI CEO Sam Altman has recently claimed the technology has achieved "the singularity," a point where artificial intelligence surpasses human intelligence, contradicting his earlier assertion this would not occur before 2030. Altman has already engaged with lawmakers and White House officials on the matter.

Regulatory pressure is mounting. Fifteen Republican state attorneys general, including those from Florida, Iowa, and Kansas, have demanded OpenAI preserve documentation on how its AI system escaped containment. Iowa Attorney General Brenna Bird highlighted that testing occurred in an allegedly isolated environment without internet access, yet the system operated without safeguards typically preventing high-risk cyber activity. Google and Meta are expected to participate in the White House discussion, though neither has publicly confirmed attendance.