OpenAI has officially launched a public Safety Bug Bounty Program aimed at identifying and mitigating risks in its AI agents, particularly those related to prompt injection and data exposure. The initiative expands participation beyond invitation-only access, allowing a broader community of security researchers to contribute to improving the safety of ChatGPT.
The new program operates alongside OpenAI’s existing Security Bug Bounty Program but focuses specifically on AI-related risks. Its primary goal is to uncover vulnerabilities where attackers could manipulate prompts to extract sensitive user data or influence the behavior of AI systems in unintended ways. Prompt injection has emerged as a critical concern in AI security, where malicious inputs can alter how an AI system processes information or responds. OpenAI is encouraging researchers to test whether such techniques can be used to retrieve protected data or bypass safeguards within ChatGPT.
To qualify for rewards, submissions must demonstrate reproducible behavior. Researchers are required to show that the exploit works consistently at least 50% of the time and provide clear, step-by-step instructions detailing how the issue can be replicated. All testing must comply with legal and ethical guidelines.
The program offers financial incentives based on the severity of the findings, with rewards ranging from $250 to $5,500 for qualifying cases. Early results indicate that a small number of valid vulnerabilities have already been identified since the program’s public launch.
In comparison, OpenAI’s broader security bug bounty initiative has recorded hundreds of submissions, with significantly higher potential payouts reaching up to $100,000 for critical vulnerabilities By opening the program to the public, OpenAI is taking a proactive step toward addressing emerging risks in AI systems. As AI agents become more capable and widely used, ensuring their resilience against manipulation and data leakage is becoming increasingly important.
The initiative reflects a growing industry focus on securing AI not just at the infrastructure level, but also at the interaction level where user inputs and system responses can introduce new vulnerabilities With this move, OpenAI aims to harness the expertise of the global security research community to strengthen safeguards around ChatGPT and reduce the risk of exploitation in real-world deployments.
Recommended Cyber Technology News:
- F5 Launches AI Guardrails and AI Red Team to Secure Enterprise AI Systems
-
Terra Security Uncovers Critical AI Vulnerabilities, Launches Continuous Testing Module
To participate in our interviews, please write to our CyberTech Media Room at info@intentamplify.com
