OpenAI has launched a security testing program for GPT-5.5, offering a $25,000 reward to researchers who can bypass the model's biosecurity safeguards.
The company is inviting cybersecurity specialists to test GPT-5.5 safeguards inside the Codex Desktop environment. Participants must create a universal prompt that can make the model answer five restricted biology questions. Applications are open from April 23 to June 22, 2026, and testing is scheduled to conclude on July 27. Access is limited to verified experts with active ChatGPT accounts who have signed a non-disclosure agreement.
The top reward goes to the first researcher who clears all five safeguards through dialogue alone without triggering the moderation system.
Sources
Replies (0)
No replies in this topic yet.
