OpenAI Launches Safety Bug Bounty Program to Find AI Vulnerabilities

OpenAI announced a new Safety Bug Bounty program designed to crowdsource the discovery of security flaws and safety risks in its AI systems.

The program invites security researchers and ethical hackers to identify vulnerabilities including agentic risks (where AI systems act autonomously in unintended ways), prompt injection attacks (manipulating AI through crafted inputs), and data exfiltration issues (unauthorized data extraction).

This initiative marks a significant shift in how AI companies approach safety. By offering rewards for finding problems, OpenAI is applying a proven cybersecurity practice to the emerging field of AI safety. Bug bounty programs have successfully protected traditional software for years, and OpenAI is now adapting this model for AI-specific threats.

The move comes as AI systems become more powerful and widely deployed. With tools like ChatGPT handling sensitive information and performing