OpenAI increases bug bounty rewards for critical product flaws
OpenAI has increased rewards in its Security Bug Bounty programme, raising the potential payout for a critical vulnerability in core products such as ChatGPT, Codex and the API Platform to as much as $30,000. Exceptional critical findings on those targets remain eligible for a base award of up to $100,000 and can now reach $172,000 when applicable bonuses are included.
The company said the changes are intended to encourage independent researchers to investigate vulnerabilities in OpenAI software that internal testing may not identify. The revised programme also introduces incentives tied to report quality, sustained participation and designated campaigns for newly launched features.
New rewards and report-quality multipliers
OpenAI said it has more than quadrupled the midpoint of reward ranges for medium-, high- and critical-severity findings on average. The exact award still depends on the target, severity and circumstances of the reported issue.
Researchers can now receive a multiplier for a high-quality submission. OpenAI defines a useful report as one with a reliable proof of concept that isolates the flaw, identifies when exploitation is possible and demonstrates the effect. The multiplier is designed to recognise the additional work that helps engineers begin remediation.
Its illustrative medium-severity example involves a read-only ChatGPT Project member uploading files without write permission. The historical award was $750; under the new calculation, a $2,000 base award combined with a 1.15 quality bonus and a 1.05 Bronze loyalty bonus produces a $2,415 payment.
Loyalty tiers and priority campaigns
The programme adds a five-tier loyalty system with scaling bonuses for researchers who repeatedly work with OpenAI. It also creates priority campaigns, which will offer dedicated bonus pools for testing newly released features during short-term, higher-reward periods.
OpenAI's examples show how the components can combine. A high-severity insecure direct object reference affecting a file in a user's Project, on the non-core Prism target, is shown increasing from a historical $1,000 award to $3,162.50 through a $2,500 base award, a 1.15 quality bonus and a 1.10 Gold loyalty bonus.
For a one-click account takeover affecting ChatGPT, OpenAI gives an illustrative new payout of $51,577: a $30,000 base award, a 1.15 quality bonus, a 1.30 priority-campaign bonus and a 1.15 bonus at the 31337 loyalty tier. An exceptional compromise of a security boundary, such as viewing arbitrary ChatGPT users' chat messages, is illustrated at $171,925 under the same quality, campaign and loyalty factors applied to a $100,000 base award.
Timing of the revised programme
The new reward ranges will apply to reports that have already been submitted but have not yet received a bounty. OpenAI also said it will pay the difference for bounties awarded during the 30 days before the announcement when the finding would have earned more under the revised base and loyalty awards.
OpenAI's security team will attend the PROMPT2PWN event in Oakland on November 20–21, where researchers can test targets in person. For businesses operating vulnerability disclosure programmes, the update underlines the practical value of setting clear severity ranges and rewarding reproducible evidence, impact analysis and long-term researcher participation.