David Sacks on Anthropic export control (twitter.com)

David Sacks recently discussed the controversial release of Anthropic's Fable model, which is based on their Mythos class but equipped with safety guardrails. However, concerns arose when a trusted partner reported a jailbreak that could potentially expose the advanced cyber capabilities of Mythos, prompting the US government's request for Anthropic to address the flaw or withdraw the model. Surprisingly, Anthropic declined to take immediate action, asserting the jailbreak was not a serious threat, which conflicted with their established reputation for prioritizing AI safety. This response led to the administration implementing export controls on Fable, an unexpected and reluctant measure, indicating the government’s frustration with Anthropic's lack of cooperation. The situation highlights significant implications for the AI/ML community, especially regarding accountability and safety measures for advanced AI systems. As Anthropic faces pressure to rectify the issue and lift the export restrictions, the incident raises critical questions about the balance between innovation, safety, and responsible governance in AI development.