OpenAI has sharply raised what it pays outside researchers for finding security flaws, and is bringing in security firm Cathedral to attack its systems alongside its own staff in a second company-wide security sprint.
Both moves were set out on 2 November, in posts on OpenAI’s next security sprint and on bigger bug bounties, and announced by @OpenAI on X.
In brief
- OpenAI says the midpoint of its medium, high and critical bounty ranges has more than quadrupled on average.
- Exceptional findings can now earn up to $172,000 with bonuses, and recent awards are being topped up.
- Cathedral researchers will use OpenAI’s models with OpenAI’s teams “to challenge our defenses”.
What changes in the OpenAI bug bounty
Under the new structure, OpenAI says the midpoint of its medium, high and critical reward ranges has more than quadrupled on average. Critical findings in ChatGPT, Codex or the API Platform can earn up to $30,000. Exceptional findings, previously capped at $100,000, can now reach up to $172,000 once bonuses are added.
Those bonuses come from three new pieces: a quality multiplier for strong reports, a five-tier loyalty programme for repeat researchers, and bonus pools for priority campaigns.
OpenAI gave worked examples of how past awards would look under the new rules, while stressing that “all scenarios and historical awards are illustrative”. In its examples, a medium finding paid $750 would now earn $2,415; a high-severity finding in Prism paid $1,000 would earn $3,162.50; a critical zero-click account takeover paid $6,500 would earn $51,577; and an exceptional finding paid $100,000 would earn $171,925.
The new ranges apply to reports not yet paid. For awards made in the previous 30 days, OpenAI says it will pay the difference, covering both the base award and any loyalty awards. Researchers who reported a flaw in the past month therefore do not lose out because the new rates arrived after their report was paid. It is also holding a researcher event, PROMPT2PWN, in Oakland on 20 and 21 November.
Cathedral and the second security sprint
The second strand is a repeat of an internal exercise. In OpenAI’s last sprint, 25% of its product engineers were dedicated to security work, using Codex. This time Cathedral’s researchers “will use OpenAI’s models alongside our teams to challenge our defenses”, the company says, arguing that “traditional cyber defenses alone are no longer sufficient”.
Gavin Kliger, Cathedral’s chief executive and co-founder, said: “OpenAI is setting an important example by partnering with outside security firms and using its own models to continuously test and strengthen its defenses. Every company should be thinking about how to put these capabilities to work. We’re excited to collaborate with OpenAI’s teams, combining their models and security expertise with our research to find weaknesses, test fixes and strengthen this approach to continuous defense.”
OpenAI also pointed to $1 billion in subsidised model access over six months for frontline defenders, offered through Daybreak customers.
What the OpenAI bug bounty rise signals
The message is that AI is changing the economics of finding bugs, and OpenAI wants more of the best finds reported to it directly. Raising the typical award, not only the headline maximum, matters most here: medium and high findings make up much of any programme, and a fourfold rise at the midpoint changes whether a researcher bothers to write one up. TSN has reported how far payouts stretched before this, including the $6,000 OpenAI paid Xint for a flaw that leaked hidden reasoning.
Cathedral is a pointed choice. TSN verified on 26 October that Cathedral’s agents had reached Anthropic systems through third-party software, so OpenAI is hiring a firm with recent, public evidence of what AI-driven attackers can do. It fits a wider pattern of OpenAI investing in its own defences, from wider agent monitoring to reported early talks to buy Trail of Bits.
What to watch is whether the sprint produces public write-ups of the weaknesses found and fixed, and whether other AI labs, several of which also run public bounty programmes, match the higher rates.
Sources
- OpenAI: Preparing our next security sprint to harden OpenAI (2 November 2026)
- OpenAI: Rewarding defenders with bigger bug bounties (2 November 2026)
- @OpenAI on X: security sprint
- @OpenAI on X: bug bounty
Related stories
- OpenAI Fixes Flaw That Leaked Hidden Reasoning, Pays Xint $6,000
- OpenAI Widens Agent Monitoring After the Hugging Face Incident
- OpenAI Is in Early Talks to Buy Security Firm Trail of Bits, The Information Reports
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
