OpenAI launches a public Safety Bug Bounty programme
The initiative targets agentic risks, prompt injection attacks, and platform integrity issues to improve resilience against emerging AI safety challenges.
OpenAI has introduced a public Safety Bug Bounty programme to identify misuse and safety risks across its AI systems. The initiative expands the company’s existing vulnerability reporting framework by focusing on harms that fall outside traditional security definitions.
The programme covers AI threats such as agentic risks, prompt injection, data exfiltration, and bypassing platform integrity controls. Researchers are encouraged to submit reproducible cases where AI systems perform harmful actions or expose sensitive information.
Unlike standard security reports, the initiative accepts safety issues that pose real-world risk, even if they are not classified as technical vulnerabilities. Dedicated safety and security teams will assess submissions and may be reassigned depending on relevance.
The scheme is open to external researchers and ethical hackers to strengthen AI safety through broader collaboration. OpenAI says the approach is intended to improve resilience against evolving misuse as AI systems become more advanced.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!