Event ·Multi-source

OpenAI 发布 GPT-5.6 Sol，对抗 prompt 注入失败率降低 6 倍

OpenAI 通过对抗训练（对抗 GPT-Red）提升了 GPT-5.6 的安全性。测试中重放了 GPT-Red 的最强攻击（均未在训练中出现），GPT-5.6 Sol 成为最鲁棒的模型。与四个月前的最佳生产模型相比，失败次数减少至 6 倍。该结果表明对抗训练可显著增强 LLM 对 prompt 注入的防御能力。

Current read

OpenAI 刚晒出 GPT-5.6 Sol，抗 prompt 注入比四个月前强了 6 倍，安全能力大幅提升。

11 sources63° AI heatLast updated 2026/7/15 17:34:30

Evidence

11 sources

Primary sourceOpenAI

2026/7/15 17:34:30View source

Related source 1elvis

2026/7/15 17:51:11View source

Related source 2Decoder

2026/7/15 19:47:53View source

Related source 3IT之家

2026/7/16 02:51:14View source

Related source 4AWS Machine Learning Blog

2026/7/13 21:01:20View source

Related source 5Greg Brockman

2026/7/14 03:56:16View source

Related source 6AI Will

2026/7/14 09:11:33View source

Related source 7Sam Altman

2026/7/14 14:26:24View source

Related source 8@OpenAIDevs

2026/7/14 18:48:15View source

Related source 9techcrunch

2026/7/14 21:50:11View source

Related source 10marktechpost

2026/7/16 18:48:04View source

Conflict 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.

Open the story