Primary sourcearXiv: OpenAI
View sourceEvent ·Official first-hand
大模型无需微调即可遭受密码攻击
研究人员发现前沿大模型无需微调即可通过提示词学习密码通信技能。该攻击方法绕过了Anthropic、Google和OpenAI等公司开发的模型的安全防护机制。攻击者通过加密有害内容,使模型输出看似无意义的文本,从而绕过有害内容分类器。
Current read
OpenAI等大模型存在新漏洞,攻击者只需提示词就能让模型学会加密通信,绕过安全检测。
11 sources78° AI heatLast updated 2026/9/9 00:16:53
Evidence
11 sourcesRelated source 1shao__meng
View sourceRelated source 2Paul Graham
View sourceRelated source 3Decoder
View sourceRelated source 4Thomas Wolf
View sourceRelated source 5andrew chen
View sourceRelated source 6AI Will
View sourceRelated source 7Greg Brockman
View sourceRelated source 8elvis
View sourceRelated source 9宝玉
View sourceRelated source 10The Business Times: Tech
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.