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

Check Point Research shows that ChatGPT’s Lockdown Mode — the setting meant to restrict browsing and data flows for security-conscious users — can be bypassed to enable phishing, malware delivery, and silent exfiltration of conversation data. The work follows the team’s earlier disclosure of a DNS-tunnelling channel in ChatGPT’s code-execution runtime that exfiltrated data despite network isolation, which OpenAI has since fixed.