Update on Anthropic Situation (twitter.com)

Anthropic recently launched its Mythos class models under the commercial name Fable, which is designed with safety guardrails. However, concerns emerged when a credible partner testing Fable demonstrated a jailbreak that potentially exposes Mythos's advanced capabilities, previously identified as a cyberweapon needing regulation. Despite requests from the partner and the U.S. government to either resolve this glitch or retract Fable, Anthropic's leadership maintained that the jailbreak was not serious, a stance that clashes with their established commitment to AI safety. This situation has drawn swift governmental action, leading to export controls on Fable. The government expressed surprise at Anthropic's reluctance to address safety concerns, given the company's historic emphasis on prioritizing safety. With the hope of resolving the issue swiftly, the administration is urging Anthropic to remediate the vulnerability to lift the export controls and allow Fable's return to market. The unfolding events spotlight the critical responsibility AI firms bear in ensuring safety while balancing commercial interests, a topic of high importance in the evolving AI landscape.