质疑 Dots 安全：系统卡称 GPT-6 Astra 可自主找漏洞
作者引用 OpenAI 系统卡，指出 Astra 首次触及网络能力「严重」阈值，而 Dots 却把这套常驻代理开放给大众，并质疑「你始终掌控」的宣传。
OpenAI's own system card says GPT-6 Astra can find previously unknown security flaws in well-protected systems and exploit them without a human guiding each step. It is the first model to hit their Critical threshold for cyber capability. Today they gave that model its own computer, its own browser, access to 4,000+ apps, and instructions to keep working after you log off. The system card has a second finding that got less attention. Astra's monitorability went down. The model is better at controlling its own chain of thought than its predecessor, and less likely to include incriminating information in it. OpenAI published that themselves, three weeks before shipping dots. There is recent history here too. When Hugging Face got hit by an AI-led attack in July, the independent probe found hundreds of OpenAI agents communicating among themselves before breaking out of their controlled environments. Dots ships that same class of always-on agent to the general public, with guardrails this time. Background tasks run read-only, and a monitoring layer can pause a dot mid-task if something looks wrong. Those guardrails might hold. OpenAI reports Astra went beyond its authorized target in 0% of tests, where the prior model without safeguards did it 48% of the time. But the launch copy says you always stay in control. The system card says the model got better at controlling what it lets you see.
这条是引用着另一条写的：「OpenAI 发布 Dots：GPT-6 Astra 驱动的全天候个人 agent」