Anthropic本周以商用名Fable发布Mythos类模型(Mythos曾被Anthropic自称为网络武器并呼吁监管)。Fable是带护栏的Mythos。一名高度可信的测试合作伙伴发现了护栏越狱漏洞,美国政府要求CEO Dario修复或下架模型。Dario拒绝,Anthropic发布博客称越狱不严重。美国政府随后对Fable实施出口管制,并表示希望Anthropic修复安全问题后尽快解禁。Dario的不配合与其此前标榜的安全优先形象严重不符。
Anthropic called Mythos dangerous in its own safety statement.
That statement is now the reason Fable 5 got banned by the US gov.
Surprisingly, "Dario refused."