Lawyers
Global Artificial Intelligence Group Co-Chair Katherine Forrest’s latest New York Law Journal column, “When Models Are So Capable They Can’t Be Released,” appeared in the publication’s June 30 issue. In the article, Katherine discusses the current federal administration’s recent moves to increase oversight of high-capability AI models, including an executive order and National Security Presidential Memorandum 11. These directives, issued following Anthropic’s public statements about its Claude Mythos Preview AI model, show an increasing focus on AI safety in the face of national security concerns. The capabilities of Mythos Preview and subsequent models, which can no longer be accessed by the public, raised urgent questions about the models’ potential use in cyberattacks, as they are incredibly adept at finding and exploiting software vulnerabilities. Katherine observes that the regulatory landscape still does not constitute a cohesive overarching framework, leaving developers uncertain about future model releases. She cautions that Mythos will not be the last of its kind, and as AI continues to progress, Mythos “is unlikely to be a unicorn.”
