Bug bounty rewards: OpenAI offers up to $172,000

An exceptional critical vulnerability in ChatGPT, Codex, or the API Platform now pays up to $172,000 in bug bounty rewards with bonuses. The program combines higher base ranges with incentives for report quality, continued participation, and priority campaigns. You need to separate that headline ceiling from ordinary awards—and check whether an existing report qualifies for the revised rates.

For independent researchers, the change puts real money behind making a finding easier to reproduce and assess. OpenAI's announcement says a reliable proof of concept should isolate the bug, establish exploitation conditions, and demonstrate impact. Those details now help a report qualify for a multiplier, not just support the underlying finding. The original write-up on openai.com outlines the changes and points researchers to the full security bug bounty program for complete details.

How OpenAI's bug bounty rewards have changed

OpenAI says it has more than quadrupled the midpoints of its reward ranges for medium, high, and critical findings—on average. That's an average. It doesn't mean every submission sees the same increase.

A critical vulnerability in a core product—ChatGPT, Codex, or the API Platform—tops out at $30,000. Exceptional critical findings on those same targets still start at a base of up to $100,000, with bonuses pushing the advertised maximum to $172,000.

That largest figure is an exceptional award with bonuses, not the standard critical payout. A $30,000 critical finding and an exceptional security-boundary compromise are different propositions in OpenAI's own examples.

The additional incentives reward different things. Quality multipliers recognize useful reporting work. Loyalty bonuses recognize continued participation. Priority campaigns direct attention toward newly launched features through short-term campaigns with dedicated bonus pools. The announcement describes what those campaigns will offer; it doesn't name an active one or its dates.

How the illustrative payouts combine bonuses

OpenAI provides the following examples to show how base awards and multipliers can interact; every scenario and historical award in this table is illustrative.

Eligibility for a bonus isn't a guarantee of that bonus. The useful distinction is between a higher starting award and extra multipliers. In the exceptional example, the illustrated base stays at $100,000; the extra comes from bonuses. In other examples, the starting award itself rises before multipliers apply.

These aren't disclosures of actual incidents or payment records. The examples don't establish that every report qualifies for every multiplier. They explain the structure of the revised bug bounty rewards, not what a researcher should expect to earn.

The headline ceiling is less precise than the exceptional example: OpenAI advertises up to $172,000, but the displayed calculation yields $171,925. Treat the ceiling as the program's advertised maximum, not the exact result of that example.

Prepare an OpenAI report for the new bonuses

The most actionable change is the quality incentive: OpenAI explicitly values evidence that helps it understand and fix the flaw. Build the submission around reproducibility and demonstrated impact.

A practical preparation sequence:

1. Check the target's classification before estimating an award. OpenAI distinguishes core and non-core assets in its examples and points researchers to the full security bounty program for complete details. Confirm the relevant program terms rather than assuming that a ChatGPT example also establishes the rate for another target.
2. Reduce the proof of concept to the behaviour that demonstrates the bug. OpenAI specifically values a reliable demonstration that isolates the vulnerability. Keep the explanation focused on what the report actually establishes.
3. Document the exploitation conditions and the resulting impact. For a permission-related finding like the illustrative Project upload scenario, explain the permission the member has and the action that nevertheless succeeds. The aim is to make the boundary failure understandable, not merely to show an unexpected result.
4. Check potential bonuses separately from the base award. Establish whether the report could qualify for a quality multiplier, which loyalty terms apply and whether a relevant priority campaign exists. Do not treat the combined multipliers in an example as automatic entitlements.

This sequence follows OpenAI's distinction between finding a weakness and producing a report that gives its engineers a head start on a fix. It also keeps the payout estimate tied to evidence: demonstrated impact belongs in the report before the hoped-for reward.

Which earlier reports get the higher rates?

Researchers with existing submissions have another reason to review the announcement: OpenAI is applying revised ranges to more than just new reports. The treatment depends on whether a bounty has already been awarded.

• For a submitted report that has not yet received a bounty award, OpenAI says it will apply the new reward ranges.
• For a bounty awarded during the 30 days before the announcement, OpenAI says it will pay the difference if the finding would have earned more under the new ranges, using the base award and loyalty awards.

The second rule is narrower than recalculating every recent award with every multiplier. OpenAI specifically names base and loyalty awards for that retrospective adjustment. It doesn't promise to add quality and priority campaign multipliers to already-awarded bounties.

That distinction determines which policy applies before any revised estimate.

The announcement text doesn't include a publication date, so the 30-day window can't be converted to calendar dates here. It's measured backward from the announcement, not from when you read this. For existing participants, that timing rule may be more useful than the maximum rewards advertised.

The announcement is not a complete payout schedule

The announcement explains the incentive structure but doesn't supply every reward range, loyalty threshold, or campaign condition. It points researchers to the full security bounty program for those. The example multipliers shouldn't be treated as a universal schedule for every target and submission.

OpenAI says high-quality reports will be eligible for a multiplier. The examples show possible combinations, but the announcement doesn't establish that any particular new report will receive them.

I wouldn't use the headline ceiling as a forecast of likely earnings. The practical case for participating is clearer when you have a finding on a relevant target, can demonstrate its impact, and can produce the reliable proof of concept OpenAI describes.

For developers using ChatGPT, Codex, or the API Platform without doing security research, this announcement contains no software update or configuration change to apply. It changes incentives for independent researchers to report weaknesses OpenAI's testing may have missed; it doesn't announce a named flaw has been fixed.

If you have a pending report with no bounty awarded, the new reward ranges apply regardless of when you submitted it. For an already-awarded bounty, check the award date against the announcement: the retrospective adjustment covers awards made in the preceding 30 days.