Anthropic hails the news, but cautions that recent events have highlighted the need for a «consistent way to assess and fix potential «jailbreaks,»» after two weeks of grueling exchanges with the administration.
Commerce secretary Howard Lutnick said on X that they had «worked closely with Anthropic to […] ensure alignment across the US Government and strengthen America’s leadership in AI,» Axios reports.
The Fable 5 model became somewhat of a joke in the AI community for having such strong safeguards that it would not answer anything even remotely related to security or biology, but Amazon engineers got it to output code for an exploit, leading to the export ban on June 12.
This behavior is now blocked 99% of the time, and Anthropic has agreed to even further safeguards on the model, pledging to work even more closely with the government in the future, including on vetting pre-release models.
Read more: Anthropic’s announcement and X post, Lutnick’s letter, Axios, Politico, CNBC.