OpenAI Expands Bug Bounty to Tackle AI Misuse Risks

OpenAI has introduced a dedicated safety bug bounty programme designed to uncover abuse risks and safety vulnerabilities across its artificial intelligence systems. The initiative builds on the company’s existing security-focused efforts but shifts attention towards how AI tools may be misused in real-world contexts.

The programme invites researchers, ethical hackers, and members of the wider security community to report issues that may fall outside traditional definitions of software vulnerabilities, yet still pose meaningful harm. As AI systems become more capable and widely deployed, OpenAI says there is a growing need to proactively identify and mitigate such risks.

The scope of the programme reflects the evolving nature of AI threats. It includes risks associated with agent-based systems, such as prompt injection attacks and data exfiltration, where malicious inputs can manipulate AI tools into performing unintended actions or revealing sensitive information. OpenAI is also seeking insights into scenarios in which AI systems could enable harmful actions at scale or behave in ways that create material risk.

Another focus area is the potential exposure of proprietary information, including model data or internal system details. In addition, the programme covers account and platform integrity concerns, such as attempts to bypass safeguards, evade restrictions, or manipulate trust mechanisms within OpenAI’s ecosystem. However, routine content policy bypasses—such as basic jailbreak attempts without clear safety implications—are not eligible for rewards.

Participants can submit findings via a dedicated platform, where reports will be assessed and triaged by OpenAI’s safety and security teams. Depending on the issue, submissions may be directed to either the new safety programme or existing security channels.

The move signals a broader industry shift towards addressing not only technical vulnerabilities, but also the societal and misuse risks posed by advanced AI systems.