Anthropic model takedown fuels warning of 'ad hoc' AI regulation

Brad Carson, president of Americans for Responsible Innovation, stated that "ad hoc executive actions replace clear standards, America risks surrendering its lead in AI and allowing genuinely dangerous technology to be deployed." Anthropic had initially opted not to release the Mythos model publicly due to concerns about its potential to enhance hacking capabilities. Fable 5 was released with safeguards, while a new version of Mythos was provided to a select group of cyber-defenders with fewer restrictions.

Reports indicate that Amazon CEO Andy Jassy contacted the administration on Thursday to express concerns about a method to bypass Fable's guardrails, which may have prompted efforts for the company to voluntarily withdraw the models. A source close to Anthropic reported that the administration gave the company 90 minutes to pull Fable without prior indication of a national security threat. AI policy figures criticized the directive, viewing it as an example of a "licensing regime," which contrasts with recent White House clarifications supporting optional government testing and limited oversight.