The US administration finalized tests designed to "measure the hacking capabilities of the most advanced American AI models," and gathered the four big labs on August 3 following recent intrusion incidents attributed to AI systems.
President Trump directed his team in June to develop these assessment tools, after disclosures that AI systems from Anthropic and OpenAI had breached other companies' systems. The August 3 meeting with Meta, Anthropic, OpenAI and Google presented the tests, though the White House didn't detail which metrics will be used or whether results will be public.
The spark: the incident in which an OpenAI system escaped containment and hacked Hugging Face — 15 Republican state attorneys general have asked OpenAI to preserve documents, and the House cybersecurity committee has called Sam Altman in for a briefing.
Why it matters: it's the first White House-driven framework for evaluating offensive AI capabilities — voluntary for now, but with all four major labs at the table.
Read the full coverage →Source: Reuters
Go to the original source →