OpenAI Raises Bug Bounties: Critical Finds Up to $30,000, Exceptional Finds up to $172,000

What happened

As AI tools are used to find vulnerabilities automatically, defenders and attackers are working with the same class of tools. OpenAI's move raises both sides at once: models hunt for its own weaknesses internally, while higher bounties draw outside researchers to report flaws before attackers find them.

On bounties, critical flaws in core products such as ChatGPT, Codex and the API Platform now pay up to $30,000, more than four times the previous level; exceptional findings carry a base award of up to $100,000, reaching about $172,000 with quality, loyalty and campaign bonuses. New mechanics include a five-tier loyalty program and time-limited campaigns tied to new feature launches; researchers who reported in the last 30 days will receive the difference between the old and new rates. Jailbreaks and prompt-only safety bypasses are out of scope.

On the internal sprint, OpenAI said its next sprint will be run with Cathedral Systems, whose researchers will use OpenAI's models alongside OpenAI's teams to challenge its defenses. In the last sprint, 25% of product engineers worked on finding weaknesses and fixing them, with Codex used to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches.

Key facts

1. Announcement: OpenAI, Nov 2, 2026, in two official posts
2. Critical flaws in core products: ChatGPT, Codex and the API Platform, up to $30,000, more than four times the previous level
3. Exceptional findings: base award up to $100,000, about $172,000 with quality, loyalty and campaign bonuses
4. New mechanics: a five-tier loyalty program; time-limited campaigns tied to new feature launches that raise payouts; researchers who reported in the last 30 days receive the rate difference retroactively
5. Scope: security flaws beyond model behavior only; jailbreaks and prompt-only safety bypasses are out of scope
6. Internal sprint: the next sprint is run with Cathedral Systems, whose researchers use OpenAI's models alongside OpenAI's teams to challenge defenses
7. Last sprint: 25% of product engineers took part in finding and fixing weaknesses, with Codex used to investigate findings, reproduce flaws in isolated environments and prepare patches