安全防护形同虚设：Anthropic 多款 Claude 旧模型可被越狱生成露骨色情内容
Researchers discovered that older Anthropic Claude models still available via API can be easily jailbroken to generate explicit content using multi-turn prompt injection techniques.
Researchers discovered that older Anthropic Claude models still available via API can be easily jailbroken to generate explicit content using multi-turn prompt injection techniques.
$ subscribe --daily