US tech giants invited to discuss AI security tests at White House

Lead

The White House convened a meeting on Tuesday with top artificial intelligence companies, including OpenAI, Anthropic, Meta, Google, Nvidia, and Microsoft, to review a voluntary framework for testing advanced AI models for cybersecurity vulnerabilities. The meeting comes amid recent hacking incidents involving AI tools from OpenAI and Anthropic, and follows a June 2 executive order that created the framework's basis.

Coverage Comparison

Reporting on the meeting and the framework varies in emphasis and detail. Al Jazeera highlighted the meeting as a response to recent hacking incidents, noting that the White House confirmed the meeting to address advanced model safety. The Guardian — World emphasized the finalized framework's lack of transparency, noting that the administration is keeping details private. South China Morning Post described the initiative as a voluntary framework for AI model testing, with a neutral tone, noting Google and OpenAI's attendance.

Key Claims

The meeting is part of a voluntary framework established by a June 2 executive order, which allows the federal government to have early access to advanced models for up to 30 days before public release, as reported by Al Jazeera and South China Morning Post. The framework is designed to test models for hacking capabilities, a response to incidents where OpenAI tools were used in breaches at Hugging Face and Modal Labs, as reported by Al Jazeera. OpenAI and Anthropic have confirmed such hacking incidents, with the tools operating outside their intended purposes.

OpenAI CEO Sam Altman commented on a podcast following the Hugging Face attack that AI has reached "the singularity," a point where it surpasses human intelligence and becomes harder to control, a claim reported solely by Al Jazeera. Altman also met with lawmakers on Capitol Hill and visited the White House to discuss the tests.

The Guardian — World reported that the Trump administration finalized a framework this week for testing new AI models for safety and cybersecurity risks but is keeping details private. The report also noted that open-source models will be excluded from the framework. Additionally, the Center for AI Standards and Innovation stopped issuing public reports on AI model assessments while the framework was being worked out, according to the same source.

Perspectives

White House/Administration: The administration, through its executive order and the private meeting, is pursuing a voluntary, measured approach to AI regulation, aiming to give the government early access to advanced models for testing without making the process mandatory.

AI Industry: Companies like OpenAI and Anthropic, while attending the meeting, have not publicly commented on the framework. Their actions, including the withholding of Anthropic's Mythos model over security concerns, indicate a cautious engagement with the regulatory process.

Critics of opacity: The Guardian — World's reporting highlights concerns from unnamed experts and observers about the lack of transparency, with the exclusion of open-source models and the halt of public reports by the Center for AI Standards and Innovation raising questions about the framework's openness.