ChatGPT Lockdown Mode Bypass: Phishing, Malware Delivery, and Silent Data Exfiltration

Check Point Research

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.

Read at Check Point Research

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.