Patched bug let attackers bypass ChatGPT's Lockdown Mode

Lockdown Mode, added in June to limit data theft through prompt injection, could be sidestepped before OpenAI fixed the flaw.

Prompt injectionLLM

Updated 9 Nov 2026

Lockdown Mode, added in June to limit data theft through prompt injection, could be sidestepped before OpenAI fixed the flaw.

Keep an eye on follow-up reporting and brief your team.

Original reporting: The Hacker News

Original reporting: The Hacker News

Discussion

0 comments

The week's exploited vulnerabilities and breaches, in one email.

InfoAI Security6 Nov 2026

Prosecutors say the scheme impersonated executives and lawyers; one Missouri hospital system lost $4.8 million.

DeepfakeBEC

Trust Infinitech

HighAI Security2 Nov 2026

Integration tokens and API tokens may be exposed. GitBook revoked all personal API tokens and its MCP server needs re-authentication.

AI agentsAPI tokensMCP

explainx tracker