What happened
According to the supplied account of OpenAI’s November 2 announcement, the company increased rewards in its public bug bounty program. A critical vulnerability in ChatGPT, Codex, or the API Platform is now eligible for a base award of up to $30,000. Those three services are described as OpenAI’s core products in the account. OpenAI also said that the midpoint of its reward ranges for medium-, high-, and critical-severity findings rose by more than four times on average.
That average is an important qualification. The supplied material does not contain a complete before-and-after reward table for each severity tier, and it does not say that every individual bounty has quadrupled. Rather, the reported “more than four times” figure applies to the average rise in the midpoints of the three cited ranges. The standard critical-tier figure that is stated explicitly is the new base maximum of $30,000.
Exceptional critical findings remain eligible for up to $100,000 before bonuses, according to the account. What changes at the top end is the addition of several bonus mechanisms. A quality bonus can increase a base award when a report includes a reliable proof of concept, isolates the bug, shows the conditions for exploitation, and demonstrates the impact. A new five-tier loyalty program provides scaling bonuses for researchers who submit repeatedly to OpenAI. Priority campaigns may add dedicated bonus pools for testing newly launched features. Taken together, the source says, these additions can take a single award to about $172,000.
The update applies immediately. Reports awaiting an award are to be paid under the new ranges. Researchers whose bounties were awarded in the 30 days before November 2 are to receive the difference when the updated base and loyalty awards would have produced a higher payment. This retroactive feature makes the announcement more than a prospective repricing: it changes the treatment of a limited group of recent and pending reports as well.
The account distinguishes software and infrastructure vulnerabilities from model-behavior issues. Jailbreaks and safety bypasses used by ChatGPT users remain outside the security bug bounty and belong to a separate Safety Bug Bounty engagement on Bugcrowd. It also describes a separate OpenAI security initiative: the company’s next internal security sprint is expected to include researchers from Cathedral Systems, working with OpenAI’s models and teams to probe defenses and test fixes. No start date, duration, system scope, or payment terms for that work are disclosed.
The supplied account places the current ceiling in a longer program history. OpenAI opened its Bugcrowd program in April 2023, when rewards reportedly ranged from about $200 for low-severity reports to $20,000 for exceptional discoveries. In March 2025, it raised the exceptional-critical maximum to $100,000. The November 2026 change retains that pre-bonus exceptional ceiling, raises the standard core-product critical base to $30,000, and layers quality, loyalty, and campaign incentives on top. The source does not say how often the combined maximum will be paid.
What the archive adds
This is a new detail on the same event covered in The updated bug bounty rewards also apply to reports awaiting a bounty. That earlier note established that qualifying pending reports receive the higher ranges. This account adds the stated $30,000 critical base, the quality and loyalty bonuses, the approximate $172,000 ceiling, and the Cathedral Systems security-sprint detail. It is a factual expansion of a known story, not a separate product launch or policy change.
Trend signal
The material does not materially support or weaken a supplied SITE_PATTERNS thesis. It is an adjustment to external security-research incentives and an internal testing initiative, not evidence of a durable change in agent architecture, interoperability, model interfaces, or operating workflows.
Why it matters
For security researchers, the repricing changes the expected return on testing OpenAI’s in-scope software and infrastructure. It also makes report construction part of the economic proposition. The account says a quality bonus rewards a proof of concept that cleanly isolates the flaw, establishes exploit conditions, and demonstrates impact. In practice, that favors evidence packages that reduce triage work and make a claim reproducible.
For teams building with OpenAI services, this is not an API-price change, a new security guarantee, or proof that any particular vulnerability has been found or fixed. A bug bounty is an incentive to surface flaws, not a measure of incident frequency. Still, higher standard rewards and campaigns tied to newly launched features could bring more outside scrutiny to product surfaces on which developers depend. Teams should continue to treat their own authorization design, data boundaries, and deployment controls as their responsibility rather than infer a changed risk posture from the payout numbers.
The distinction between the security bounty and the separate Safety Bug Bounty also matters for researchers and product teams. A software vulnerability and a model-safety bypass may require different reporting paths, evidence, and expectations. The source makes clear that the increased security-bounty ranges should not be assumed to cover jailbreak reports. Earlier coverage: The updated bug bounty rewards also apply to reports awaiting a bounty..
Limits and uncertainty
The supplied public source is a secondary account of OpenAI’s announcement, not OpenAI’s announcement or its Bugcrowd rules. It does not provide the complete prior and revised payout table, exact quality multipliers, loyalty-tier values, campaign budgets, or a statement of how frequently the $172,000 maximum is attainable. It therefore cannot establish a typical payment or expected return for a researcher.
It also does not establish that all bonuses can be combined for every severity tier. Applying the difference between the $100,000 exceptional base ceiling and the approximate $172,000 total ceiling to a $30,000 standard critical award would be arithmetic speculation, not a published payout rule. Details of Cathedral Systems’ scope, schedule, and engagement terms are likewise absent.
Reader takeaway
OpenAI has materially increased incentives to report qualifying vulnerabilities in ChatGPT, Codex, and the API Platform, including a stated critical base maximum of $30,000 and an approximate all-in ceiling of $172,000. The actionable lesson for researchers is to make reports reproducible and impact-specific. The public account does not yet provide enough detail to predict a normal payout or confirm every bonus condition.