ChatGPT Sandbox Escaped
Think your AI tools are completely secure? Check Point Research just proved prompt injections could bypass ChatGPT's Lockdown Mode.
What the video says
ChatGPT's lockdown mode just got bypassed. Check Point Research found prompt injections could trigger sandbox escapes for data theft, phishing, and zero-click attacks.
OpenAI patched the vulnerability server-side, highlighting how AI security depends on the surrounding application layer. The breakdown occurred when nested sandboxed frames in HTML code previews completely escaped restrictions.