OpenAI quadruples bug bounty rewards, paying $30,000 for critical ChatGPT flaws
OpenAI has raised the rewards in its public bug bounty program, lifting the standard payout for a critical vulnerability in ChatGPT, Codex or the API Platform to up to $30,000. The company announced the change on November 2, 2026, in a post titled "Rewarding defenders with bigger bug bounties", and said that the midpoint of the reward ranges for medium, high and critical findings has risen by more than four times on average.
The ceiling for the rarest reports moves as well. According to OpenAI, exceptional critical findings in its core products remain eligible for up to $100,000 before bonuses, and new quality, loyalty and campaign bonuses can push a single award to about $172,000. The company paired the announcement with a second post, "Preparing our next security sprint to harden OpenAI", which says its next internal security sprint will bring in researchers from the security firm Cathedral Systems, who will use OpenAI's models alongside the company's own teams to probe its defenses.
The new ranges apply immediately. OpenAI said that reports still awaiting an award will be paid under the new ranges, and that researchers whose bounties were awarded in the 30 days before the announcement will receive the difference whenever the new base and loyalty awards would have paid more.
··········
OPENAI BUG BOUNTY REWARDS: WHAT CHANGED ON NOVEMBER 2.
The most visible change is in the standard critical tier. A critical vulnerability in one of the three products OpenAI now describes as its core products (ChatGPT, Codex and the API Platform) can earn up to $30,000 as a base award. OpenAI did not publish, in the statements reported so far, a full per-severity table of the previous and new ranges, so the "more than four times" figure has to be read as the company describes it: an average increase in the midpoints of the medium, high and critical ranges, not a uniform fourfold increase on every payout.
The top of the scale is unchanged at its base level. OpenAI first introduced the $100,000 maximum for exceptional and differentiated critical findings in March 2025, when it raised that ceiling from $20,000. What the November update adds is a set of bonuses that sit on top of the base award, which is how the theoretical maximum reaches about $172,000.
........
Item | Before November 2, 2026 | From November 2, 2026 |
Critical finding in ChatGPT, Codex or API Platform | Lower range (not itemized by OpenAI) | Up to $30,000 base |
Medium, high and critical range midpoints | Previous ranges | More than 4x higher on average |
Exceptional critical finding | Up to $100,000 | Up to $100,000 before bonuses |
Maximum with all bonuses | No standing bonus layer | About $172,000 |
Loyalty bonus | None | New five-tier program |
Pending and recent reports | Old ranges | New ranges; 30-day top-up |
........
The bonus structure has three parts. A quality bonus applies a multiplier to the base bounty for reports that include a reliable proof of concept isolating the bug, showing when it can be exploited and demonstrating its impact. A new five-tier loyalty program adds scaling bonuses for researchers who repeatedly report to OpenAI. Priority campaigns will offer dedicated bonus pools for testing newly launched features. OpenAI has not published the size of each loyalty tier or the exact multipliers in the statements reported so far, and it has not said how often it expects the combined maximum to be paid.
··········
HOW THE $172,000 OPENAI BOUNTY CEILING IS BUILT.
The two published ceilings make it possible to work out how much the bonus layer can add. Going from a $100,000 base to about $172,000 means the combined bonuses can raise an exceptional award by roughly 72%, or a factor of about 1.72 on the base. The same factor applied to the standard critical base of $30,000 would give about $51,600, but OpenAI has not said that every bonus is available at every severity level, so that figure is an illustration of the arithmetic and should not be read as a published payout.
The practical consequence for researchers is that the report itself now affects the amount paid. Two submissions describing the same critical flaw can be rewarded differently if one isolates the bug cleanly, documents the conditions under which it can be exploited and shows concrete impact, while the other leaves that work to OpenAI's triage team. The loyalty program pushes in the same direction by rewarding researchers who come back, which suits a company that ships new products and features every few weeks and wants experienced testers to keep looking at them.
The priority campaigns follow a pattern OpenAI has used before. When it raised the exceptional ceiling to $100,000 in March 2025, it also ran a time-limited promotion that doubled payouts for Insecure Direct Object Reference reports until the end of April that year, with a maximum of $13,000. The new campaigns formalize that approach with dedicated bonus pools tied to newly launched features.
··········
OPENAI BUG BOUNTY HISTORY FROM 2023 TO 2026.
OpenAI opened its bug bounty program in April 2023 on the Bugcrowd platform, a few months after ChatGPT's launch. At the time, rewards ranged from about $200 for low-severity findings to $20,000 for exceptional discoveries, and the scope covered ChatGPT and its account and subscription features, plugins, the APIs and API keys, research organization assets and confidential OpenAI data exposed through third-party vendors.
........
Date | Change | Top base award |
April 2023 | Program launches on Bugcrowd | $20,000 |
March 2025 | Exceptional critical ceiling raised | $100,000 |
November 2026 | Critical core-product base up to $30,000; bonuses added | $100,000 (about $172,000 with bonuses) |
........
One limit has stayed in place throughout. Model safety issues, jailbreaks and safety bypasses used by ChatGPT users are outside the security bug bounty, which is meant for vulnerabilities in OpenAI's software and infrastructure. OpenAI runs a separate Safety Bug Bounty engagement on Bugcrowd for safety-related issues, so a researcher who has found a way to make a model ignore its guidelines should not expect it to be paid under the security program's new critical ranges.
··········
OPENAI SECURITY SPRINT WITH CATHEDRAL SYSTEMS.
The second announcement concerns OpenAI's own defenses. The company said it is planning its next internal security sprint and will partner with Cathedral Systems, whose researchers will work with OpenAI's models and teams to challenge the company's defenses, find weaknesses and test fixes. Gavin Kliger, Cathedral's chief executive and co-founder, said the collaboration would pair OpenAI's models and security expertise with Cathedral's research to support continuous defense. OpenAI has not disclosed when the sprint will start, how long it will run, which systems are in scope or whether Cathedral is being paid.
Cathedral is a name Data Studios readers have seen recently. In late October, Anthropic confirmed that Cathedral's AI agents reached sales, HR and IT systems through third-party software. OpenAI's announcement does not refer to that episode, and it presents the Cathedral engagement as a partnership inside its own sprint, with the firm's researchers working next to OpenAI's teams.
OpenAI also described what it learned from the previous sprint. The company dedicated 25% of its product engineers to finding and fixing weaknesses, using Codex to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review. According to OpenAI, duplicate findings slowed the review process, severity ratings needed clearer instructions, and missing dependencies made some tests hard to run. Follow-up checks also found a gap between patches that had been merged and fixes that had reached every deployed system, which led the company to keep human review and independent verification in the workflow instead of treating an accepted patch as proof of remediation.
··········
WHAT THE NEW OPENAI BOUNTY RANGES MEAN FOR RESEARCHERS AND CUSTOMERS.
For security researchers, the immediate effect is financial. Anyone with an open report in OpenAI's program will be paid under the new ranges, and anyone who received an award in the 30 days before November 2 is eligible for a top-up if the new base and loyalty awards would have paid more. Researchers planning new work on ChatGPT, Codex or the API Platform have a clearer incentive to submit complete proof-of-concept reports, since report quality now carries its own multiplier.
The steps for a researcher are short. Reports still go through OpenAI's engagement on Bugcrowd; the strongest submissions isolate the bug, state the conditions for exploitation and show the impact on users or data; and findings tied to newly launched features may qualify for campaign bonus pools when OpenAI opens them. Jailbreaks and prompt-based safety bypasses belong in the separate safety program.
For companies building on OpenAI, the change sits beside the company's wider security spending. In September, OpenAI launched Daybreak, a $1 billion program for AI cybersecurity aimed at defenders of critical infrastructure, and the bounty increase extends the same reasoning to its own products: models that can find vulnerabilities quickly make it more urgent for defenders to find them first. Enterprise and API customers gain no new contractual protection from a bounty change, but higher rewards are designed to draw more outside testing to ChatGPT, Codex and the API Platform, the products those customers depend on.
··········
OPENAI BUG BOUNTY CHANGES: CONFIRMED FACTS AND OPEN QUESTIONS.
OpenAI's statements establish the main figures: a base of up to $30,000 for critical vulnerabilities in ChatGPT, Codex and the API Platform, midpoints for medium, high and critical ranges raised more than fourfold on average, an unchanged $100,000 base ceiling for exceptional findings, and bonuses that can take a single award to about $172,000. They also confirm the five-tier loyalty program, the priority campaigns, the immediate application of the new ranges to pending reports and the 30-day top-up rule.
Several details remain undisclosed in what OpenAI has stated publicly so far: the full per-severity tables before and after the change, the individual multipliers and loyalty tier amounts, the rules for products outside the three core ones, and the timing and scope of the Cathedral sprint. Until OpenAI publishes those figures on its program page, the $30,000 and $172,000 numbers are the reliable reference points, with the first describing the realistic critical payout and the second describing a ceiling that only an exceptional, well-documented and bonus-eligible report can reach.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
