Multiple outlets report that cybersecurity researchers found a way to bypass ChatGPT’s “Lockdown Mode,” allowing access to protected chat content. The issue involved using HTML-related interactions (such as popups or HTML previews) rather than requiring the usual user interaction described as part of the protection.
TechRadar and The Hacker News both say OpenAI has since patched the vulnerability. They describe Lockdown Mode as a feature introduced to reduce data exfiltration risk from prompt injections, and frame the bypass as a method that undermines that protection in practice. Business Line also reports the work of researchers, identifying the potential for data theft and related harms.
Across coverage, the shared core is that an exploit path was disclosed and then addressed by OpenAI. The differences are mainly in emphasis: some outlets focus on the mechanics of bypassing Lockdown Mode and gaining chat history access, while others foreground the security impact and the type of downstream attacks researchers say the flaw could enable.
