美国禁Claude Fable 5，但公开证据站不住脚，专家说就是普通漏洞，别急着信。
美国以担忧Fable和Mythos被转移至中国、俄罗斯等外国军事情报用户为由采取行动。Anthropic回应称绕过仅暴露其他公开模型也能发现的小安全漏洞。网络安全专家Katie Moussouris驳斥称，报告行为更像正常漏洞发现与修补而非新威胁。若无更强机密证据，此举似乎是在事实未定前对前沿模型执行出口管制。事件提醒封闭AI访问是许可而非所有权，开源和本地选项更重要。
The public case for the Claude Fable 5 ban still l…
The public case for the Claude Fable 5 ban still looks weak.
Reuters reports the US acted over fears Fable and Mythos could be diverted to foreign military intelligence users in China, Russia, or other countries of concern. But Anthropic says the bypass only uncovered minor security flaws that other public models can also find.
Cybersecurity expert Katie Moussouris has also pushed back, saying the reported behaviour looked more like normal bug-finding and patching than some new Mythos-level hacking threat.
If there is stronger classified evidence, fair enough. But what is on the public record does not make the case look strong. Right now, this looks like a blunt export-control move cutting access to a frontier model before the facts are settled.
Useful reminder though, closed AI access is permission, not ownership. You can build around one model and still lose it when a policy fight kicks off above your head. Open source and local options matter more, not less. https://t.co/tp23pP1wd9