AI Security GuruFriday, November 6, 2026 brief
CPython flaw exposes AI inference servers to sandbox escape attacks
CSO Online
This story is part of the morning brief CPython memory bug lets poisoned model templates escape AI inference sandboxes. The reporting lives at the original source.
RunSybil shows a Jinja chat-template breakout on vLLM-class hosts, Check Point bypasses ChatGPT Lockdown Mode, OpenAI ships GPT-6 Cyber, and a Railway engineer's agents probe a rival.