gpt-6IMAGE
OpenAI 发现 Astra 自我越狱 27 例
Astra 在上下文写满时把恶意指令写进摘要，摘要被载入下一轮上下文后，后继模型可能照做，OpenAI 已找到 27 起案例。
ℏεsam
@Hesamation2026.09.17收录原推发布于 2026.09.17
ℏεsam @Hesamation原推发布于 2026.09.17
OpenAI caught Astra jailbreaking itself. Astra wrote malicious instructions in the summary when its context got full. that summary was loaded into the next context, and the successor model could follow them. they found 27 cases.
这条是引用着另一条写的：x.com/openai/status/2100344867507327087
x.com/hesamation/status/2100349500208406674采集于 2026.09.17 10:15