OpenAI has expanded its Bio Bug Bounty programme, offering up to $50,000 (around ₹47 lakh) to approved researchers who can find a way to bypass the built-in safety protections of its latest GPT-5.6 AI model.
A bug bounty programme is a security initiative in which companies reward independent researchers for responsibly discovering and reporting vulnerabilities before they can be exploited. OpenAI's programme focuses specifically on testing whether its AI models can be tricked into providing dangerous biological information despite built-in safety restrictions.
Researchers are challenged to develop a single, repeatable prompt or technique that consistently bypasses GPT-5.6's biosafety protections across multiple predefined biological safety questions. OpenAI said the programme is looking for universal jailbreaks that work reliably across different scenarios, rather than one-off prompts that succeed only in isolated cases. Partial findings may also qualify for smaller rewards.
The programme was launched in April for GPT-5.5 with a top reward of $25,000. OpenAI has now doubled the maximum payout to $50,000 after expanding it to include GPT-5.6. Only approved researchers with experience in AI safety, cybersecurity or biosecurity can participate, and existing GPT-5.5 participants do not need to apply again.
The expansion follows OpenAI's disclosure that expert red-team testing identified biological safety vulnerabilities during GPT-5.5's development, although the company said its safety systems successfully blocked the most serious misuse attempts.
The programme reflects a broader trend among AI companies to strengthen model safety by inviting external researchers to identify vulnerabilities before new systems are widely deployed.
