gpt-6POST
gpt-6-astra security agent trips OpenAI cyber risk filter
A security-analysis agent on gpt-6-astra occasionally hit OpenAI's cybersecurity risk controls; switching sensitive steps to deepseek-flash stopped the triggers without hurting quality.
Cos(余弦)😶🌫️
@evilcosArchived 6 days agoOriginally posted 2026.10.21
Cos(余弦)😶🌫️ @evilcosOriginally posted 2026.10.21
真有意思，我们的一个安全分析 Agent，走 gpt-6-astra，偶尔会触发 OpenAI 的 cybersecurity risk 风控，于是前两天我用 deepseek-flash 让其做优化工作，把安全敏感操作缓解下，就再没触发风控，分析质量也没下降。另外，实际上 deepseek-flash 在许多方面已经非常满足我的安全分析工作场景了…
x.com/evilcos/status/2112823791118389474Archived 2026.10.22 10:10