OpenAI Bug Bounty: New $30,000 Critical Tier vs Old Rates

OpenAI overhauled its Bugcrowd-run bug bounty program, more than quadrupling the average midpoint of reward ranges for medium, high and critical findings and setting a $30,000 ceiling for critical bugs in core products such as ChatGPT, Codex and the API Platform. Stacked quality, loyalty and priority-campaign multipliers can push a critical award to roughly $51,577 and the exceptional $100,000 base to about $171,925, while jailbreak reports remain unpaid. The company also announced a new internal security sprint involving outside researchers from Cathedral Systems.

The OpenAI bug bounty now pays more than its previous ranges at every severity level OpenAI has illustrated, and the new scheme wins most clearly for critical findings in core products. In a post dated November 2, 2026 https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ , OpenAI said it is "more than quadrupling the midpoint of our reward ranges for medium, high, and critical findings on average", with critical bugs in ChatGPT, Codex or the API Platform paying up to $30,000. On the same day OpenAI announced its next internal security sprint https://openai.com/index/preparing-our-next-security-sprint-to-harden-openai/ , which will involve outside researchers from Cathedral Systems. For researchers who can write a reliable proof of concept against a core product, the new scheme is clearly the better deal. For anyone hoping to be paid for jailbreaks, nothing has changed. Three things changed. The base ranges for medium, high and critical findings rose, with OpenAI describing the average increase in range midpoints as more than fourfold. A critical bug in a core product now has a $30,000 ceiling, according to OpenAI's announcement https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ . And a bonus layer now sits on top of the base award. The top end is less dramatic. The $100,000 exceptional ceiling was first introduced in March 2025 https://www.datastudios.org/post/openai-bug-bounty-rewards-30000-critical-chatgpt-codex-cathedral-security-sprint , when it was raised from $20,000. The November update keeps that base and lets bonuses push it to $172,000, a figure also confirmed by Cybernews https://cybernews.com/ai-news/openai-increases-bug-bounty-rewards/ . The program itself is run through Bugcrowd https://bugcrowd.com/engagements/openai . OpenAI published four illustrative cases in its bounty post https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ . The table below reproduces them, with that post as the source for every figure. | Finding | Historical award | New computation | New reward | |---|---|---|---| | Medium, ChatGPT: read-only member can upload files to a Project | $750 | $2,000 x 1.15 quality x 1.05 Bronze loyalty | $2,415 | | High, Prism non-core : IDOR reading a Project file | $1,000 | $2,500 x 1.15 quality x 1.10 Gold loyalty | $3,162.50 | | Critical, ChatGPT: one-click account takeover | $6,500 | $30,000 x 1.15 x 1.30 priority campaign x 1.15 31337 loyalty | $51,577 | | Exceptional, ChatGPT: viewing arbitrary users' chat messages | $100,000 | $100,000 x 1.15 x 1.30 x 1.15 | $171,925 | Source: OpenAI, "Rewarding defenders with bigger bug bounties" https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ . Reading the rows against each other, the uplift is uneven. The medium and high examples rise by about three times. The critical account takeover goes from $6,500 https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ to $51,577 https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ , close to eight times, and that is only with a priority campaign and top loyalty tier stacked. The exceptional case rises by under two times, because its base was already high. If you hunt for medium bugs on non-core products, expect a modest gain. If you find critical bugs in ChatGPT, Codex or the API, the change is large. Researchers with a track record and a clean write-up win most. The multipliers reward reports that include a reliable proof of concept, repeat participation through the loyalty tiers Bronze up to a top tier named "31337" , and timing, since priority campaigns carry dedicated bonus pools for newly launched features. One-off reporters win less. A first-time submitter gets the base range and, at best, the quality multiplier. Note also that the headline $30,000 applies to core products only; the Prism example in the table is described by OpenAI as non-core and is paid on a lower base. OpenAI says it is planning its next internal security sprint https://openai.com/index/preparing-our-next-security-sprint-to-harden-openai/ with new researchers and partners using its models to test its own systems. The previous sprint put 25% of OpenAI's product engineers onto finding and fixing weaknesses, using Codex to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review, according to the same post. Cybernews reports https://cybernews.com/ai-news/openai-increases-bug-bounty-rewards/ that Cathedral is an AI penetration-testing company founded by former DOGE staff, and names its CEO as Gavin Kliger. AI Lookout https://ailookout.news/blog/openai-security-sprint-cathedral-bug-bounty-rewards-quadrupled/ adds that OpenAI's security team is scheduled to appear at the PROMPT2PWN conference in Oakland on November 20 and 21. Model safety issues and jailbreaks are outside the security program and belong to a separate Safety Bug Bounty engagement on Bugcrowd https://www.datastudios.org/post/openai-bug-bounty-rewards-30000-critical-chatgpt-codex-cathedral-security-sprint . Several things are also unpublished. CellCog points out https://cellcog.ai/blog/openai-bug-bounty-security-sprint/ that the posts give no sprint start date and no findings count from the last sprint, and AI Lookout https://ailookout.news/blog/openai-security-sprint-cathedral-bug-bounty-rewards-quadrupled/ notes there is no target list. Treat any claim about the sprint's results as unverified until OpenAI publishes them. The structural signal is a frontier lab buying outside attack labour at consumer-grade prices while committing $1 billion in subsidised access over six months to frontline defenders through the Daybreak programme https://openai.com/index/daybreak-for-frontline-defenders/ . OpenAI's stated reasoning is that traditional cyber defences alone are no longer sufficient, and that AI should be put to work finding flaws before attackers do. Practical consequences follow. A $30,000 ceiling is modest against the cost of a one-click account takeover in a product with a very large user base, so expect competition from private buyers to remain. Teams building on agents should assume the same bug classes the table lists broken access control, IDOR, account takeover will recur in their own products. Our sandbox containment playbook https://dev.to/articles/ai-agent-sandbox-containment-playbook-2026 covers isolation for agent workloads, the Codex remote agent update https://dev.to/articles/openai-codex-remote-agent-update-2026 shows the tooling OpenAI uses on the patching side, and the computer use agents guide https://dev.to/articles/chatgpt-computer-use-agents-2026-guide maps the attack surface that newer features add. OpenAI also faces non-technical exposure, as in the Apple trade secret dispute https://dev.to/articles/apple-vs-openai-hardware-trade-secret-lawsuit , but that is a separate matter from vulnerability disclosure. Q: What does the new OpenAI bug bounty pay for critical vulnerabilities? A: A critical vulnerability in ChatGPT, Codex or the API Platform earns up to $30,000 as a base, according to OpenAI https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ . Exceptional findings keep a $100,000 base and can reach $172,000 with bonuses, as Help Net Security https://www.helpnetsecurity.com/2026/11/03/openai-bigger-bug-bounty-rewards/ also reports. Q: When does the Cathedral Systems security sprint start? A: No start date has been published. OpenAI's announcement https://openai.com/index/preparing-our-next-security-sprint-to-harden-openai/ says it is planning the sprint, and CellCog https://cellcog.ai/blog/openai-bug-bounty-security-sprint/ notes the absence of a date. Q: Are AI jailbreaks covered by the OpenAI bug bounty? A: No. Model safety issues and jailbreaks are out of scope for the security program and go to the separate Safety Bug Bounty on Bugcrowd, per DataStudios https://www.datastudios.org/post/openai-bug-bounty-rewards-30000-critical-chatgpt-codex-cathedral-security-sprint . Q: Do older reports get the new reward ranges retroactively? A: Partly. Pending unaudited reports receive the new ranges, and bounties awarded in the 30 days before the announcement are topped up if the new formula pays more, according to OpenAI https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ .