They were scheduled to meet in person with the White House on Monday, sources tell CNBC, and are now finding themselves deeply enmeshed in current politics.
Government officials are now candidly telling Axios that Anthropic «screwed» them, saying they failed to «honor» the recent Executive Order on AI — which calls for a 30-day vetting period for new models, and are calling Anthropic a «bad actor.»
There are also differing narratives in the Axios report, where on the one hand, Anthropic says they «received explicit approval» to deploy Fable, and another where the administration threatened export controls several weeks ago, fearing the model could be exploited by bad actors.
The exploit uncovered in the Fable and Mythos models are replicable in other frontier and open source models, security researchers say in an open letter, and shutting down Fable at a critical time gives an advantage to attackers over defenders. They say there is only a 90-day window until Chinese models reach the same capabilities.
Read the Axios report here, also see the security researchers’ open letter, reports on euters, CNBC and Politico.