Anthropic is headed to Washington on Tuesday to review the federal government's framework for evaluating national security risks from AI models before their public release.
The White House will host a discussion with industry partners after security officials met the president's 60-day deadline to draft a blueprint, a government official confirmed.
Anthropic, creator of the large-language model Claude, along with OpenAI and Alphabet, received invitations, The Information reported on Monday, citing people with the matter.
The host, the outlet reported, will be National Cyber Director Sean Cairncross, who is the president's principal adviser on cybersecurity policy.
Anthropic will attend the review, a source familiar with the matter confirmed to Barron's on Monday. OpenAI and Alphabet didn't respond to Barron's request for comment.
The framework is the outgrowth of a June 2 executive order that set a 2-month deadline for developing a benchmarking process to assess the advanced cyber capabilities of AI models and designating what makes a model a cyber risk.
The order also established a voluntary process that would give the federal government early access to AI models to promote "secure innovation and strengthen the cybersecurity of critical infrastructure."
Cyber threats posed by AI models came to the forefront after Anthropic found that Mythos, its most capable AI model, has strong hacking capabilities. The company has limited the release of Mythos.
Comments