OpenAI’s Safety Bug Bounty program addresses AI abuse and safety risks, aiming to enhance system security and mitigate misuse that could cause harm. This initiative complements the Security Bug Bounty by accepting reports that don’t qualify as security vulnerabilities. The program targets specific AI scenarios, such as agentic risks, where attacker-controlled text may hijack agents like ChatGPT, leading to harmful actions or exposure of sensitive information. Risks related to OpenAI’s proprietary data are also addressed, particularly concerning model outputs that reveal confidential insights. Furthermore, the program focuses on account and platform integrity vulnerabilities, including loopholes in anti-automation measures. While some areas, such as jailbreaks, are not included, OpenAI occasionally runs private bounty campaigns on specific harm types. Researchers contributing valuable findings may receive rewards, but submissions must demonstrate plausible harm and actionable solutions, excluding widely known or easily identifiable issues.
Source link
Share
Read more