OpenAI Doubles Bio Bug Bounty Rewards to $50K in Expanded Biosafety Program.

OpenAI is expanding its biosafety testing program into an ongoing initiative with doubled rewards, inviting qualified researchers to probe its frontier models for vulnerabilities that could enable misuse in biology.

The company announced that its Bio Bug Bounty program is transitioning from an ad-hoc effort to a permanent private initiative. The reward structure now offers up to $50,000 for researchers who can identify a universal jailbreak capable of defeating OpenAI's predefined biosafety challenges across its most advanced models.

The program specifically seeks researchers with backgrounds in AI red teaming, security, or biosecurity. Rather than testing individual edge cases, OpenAI is asking participants to find systematic vulnerabilities that work across multiple scenarios and model versions.

This focus on universal jailbreaks represents a shift from earlier bug bounty efforts that often centered on specific prompt injections or narrow exploitation techniques. By looking for broader patterns of failure, OpenAI aims to strengthen its safety layers at a more fundamental level.

The predefined biosafety challenge remains under wraps for program participants, but the approach suggests OpenAI is stress-testing its models against scenarios where biological knowledge could be weaponized or misused. The $50,000 payout doubles the previous maximum reward, signaling both the severity of potential risks and the difficulty of finding exploitable vulnerabilities.

Unlike public bug bounties that allow anyone to participate, this initiative operates as a private program with invitation-only access. OpenAI is specifically recruiting researchers who already work in adjacent fields, likely to ensure participants have both the technical sophistication to probe frontier models and the ethical framework to handle sensitive biosafety information responsibly.

The private structure also allows OpenAI to control the disclosure process more tightly. Given the nature of biosafety vulnerabilities, public disclosure could pose risks even as it drives improvements.

This approach mirrors strategies used in other high-stakes security domains, where coordinated disclosure and vetted participants reduce the window between vulnerability discovery and remediation. For AI systems where capabilities continue to expand rapidly, having a standing team of external red teamers provides ongoing stress-testing as new models ship.

The Bio Bug Bounty expansion fits into OpenAI's wider portfolio of safety initiatives. The company has previously launched specialized programs like Daybreak for security-focused testing, indicating a domain-specific approach to identifying risks.

Biosafety concerns have drawn particular attention as large language models become more capable at synthesizing technical knowledge. While these models can accelerate legitimate research and education, they also lower barriers to accessing information that could enable harmful applications in synthetic biology or biochemistry.

By offering substantial rewards and recruiting domain experts, OpenAI is acknowledging that internal testing alone cannot surface all potential vulnerabilities. External researchers bring different perspectives, attack strategies, and domain knowledge that complement in-house red teaming efforts.

The program details and application process are available through the link in OpenAI's announcement. Researchers interested in participating will need to demonstrate relevant expertise in AI safety, security research, or biosecurity work.