The Briefing: Anthropic’s Contradiction

Still, SpaceX was quickly supplanted in the business limelight by the latest Anthropic-government drama, this time over the AI firm’s recently released Fable 5 model. Is it just me, or is there something contradictory about a company that was ostensibly formed to “build safer systems” having a run-in with the U.S. government over the safety of its latest models? 

Anthropic’s statement here explains its view of the situation: it argues that “perfect jailbreak resistance is not currently possible for any model provider” and it believes in its strategy of “defense in depth” . And it disagrees that issues raised with the latest model required it to be pulled from release. Those close to the administration disagree. 

What makes this a multi-layered situation is that some companies, including Microsoft, were already holding off on using the model because, as we wrote here, Anthropic had introduced a new policy with Fable 5 to hold onto customer data for 30 days. Anthropic says this data retention policy was part of its cyber defense strategy although it acknowledged on Friday that it “carries real costs for us with customers.”

It’s not just the government that has tense relations with Anthropic. As we reported here, Anthropic has sharp elbows when it comes to dealing with other businesses. The bottom line is that despite Anthropic’s warm and fuzzy origin story, it has emerged as the most cut-throat competitor in tech, operating much more efficiently than its better known but more chaotically managed rival OpenAI. Will this drama give OpenAI an opening to catch up? Stay tuned for further developments.