OpenAI has increased what it will pay independent security researchers for identifying flaws in products including ChatGPT, Codex and the API Platform. The changes include additional money for strong submissions, a tiered loyalty scheme and dedicated funding for campaigns focused on recently released features.
An exceptional critical weakness can now produce a payment of up to $172,000 once all available incentives are counted. The financial lure is obvious, but the more consequential shift may be OpenAI’s growing emphasis on research that gives engineers enough evidence to reproduce, assess and repair a defect.
Higher rewards will attract attention. They also make careful testing and private reporting even more important.
What OpenAI changed
Across the medium, high and critical categories, the middle value of each payout band has risen by more than four times on average. Critical weaknesses affecting core offerings such as ChatGPT, Codex and the API Platform may receive as much as $30,000 through the standard award.
OpenAI retains a higher ceiling for unusually serious critical discoveries. Those submissions can qualify for up to $100,000 before incentives are added, while the overall amount may reach $172,000.
Submissions that meet the company’s quality standards may also have their base award increased by a multiplier. OpenAI describes a dependable proof of concept as the key to that standard because it identifies the faulty behavior, explains the conditions under which it can be triggered and establishes the consequences.
Researchers do not lose the benefit of the revision if their reports are still awaiting payment. OpenAI will review those cases under the revised bands. It will also make up the shortfall for awards issued during the 30 days preceding the announcement when the previous calculation was lower, covering both ordinary bounties and loyalty payments.
Why the larger rewards matter
The new numbers give experienced bug hunters a stronger reason to spend time on OpenAI’s products. The multiplier for strong submissions adds another reason to avoid speculative claims and instead show precisely how a vulnerability behaves and what damage it may cause.
That evidence has practical value. A clear proof of concept allows an engineering team to confirm the report, reproduce the failure and determine whether a proposed correction closes the issue. It also reduces the ambiguity that can make triage slower.
The loyalty scheme introduces five levels, with the bonuses increasing as participants move upward. Its purpose is to encourage researchers to develop a continuing relationship with the program rather than submit an isolated report and disappear. Separate pools are also planned for limited campaigns centered on newly introduced features.
There is an obvious risk on the other side of that equation. More testing can mean greater pressure on live systems, and higher payouts may encourage some participants to push boundaries. Responsible disclosure therefore remains indispensable. Accepting a bounty does not authorize access to other people’s information, service disruption or premature publication of exploit instructions.
How outside researchers fit into OpenAI’s security work
OpenAI says the bounty initiative operates alongside its Defense Factory. There, company engineers use OpenAI’s own models to locate vulnerabilities, reproduce reported behavior and check whether proposed repairs work.
Outside participation supplies a different line of sight. Internal testers can explore established patterns systematically, while independent researchers may question assumptions, uncover unusual usage paths and concentrate on overlooked edge cases.
The bounty program adds a route for bringing those findings into the company’s process. It also shifts more of the effort of proving a vulnerability’s existence toward the researcher. OpenAI benefits when that evidence is strong enough for engineers to act without extensive reconstruction.
Security research decision checklist
Before beginning an authorized test, researchers should consider the following:
- Confirm that the target and testing method fall within the responsible-disclosure program’s scope.
- Use only accounts, information and conditions that they are permitted to test.
- Record the requests, responses and side effects needed to reproduce the behavior.
- Narrow down the circumstances that trigger the suspected weakness.
- Establish the effect with a dependable proof of concept.
- Keep real customer information out of test material and exploit details out of public posts.
- Send credible findings through the responsible-disclosure channel and allow time for a correction.
What to watch next
The size of the headline award will draw attention, but the more useful measure is whether the revised scheme yields findings that OpenAI’s engineers can reproduce and resolve.
The first decisions under the new bands will show how quality multipliers and loyalty payments are being handled. Researchers will also reveal whether the tiered scheme encourages repeat participation rather than one-off submissions. Campaigns covering newly released features deserve particular scrutiny because they test products during the period when their behavior is least settled.
Researchers gain a clearer financial incentive to pursue difficult flaws and provide stronger technical evidence. OpenAI gains another way to draw on specialized scrutiny. Users benefit only when those findings lead to effective repairs.
Key takeaways
- Exceptional critical discoveries can earn as much as $172,000 after incentives are included.
- ChatGPT, Codex and the API Platform are among the core offerings covered by the revised critical-finding range.
- A dependable proof of concept helps engineers verify, assess and correct a vulnerability.
- The bounty initiative supports OpenAI’s Defense Factory rather than replacing internal testing.
- Careful testing and confidential reporting remain essential despite the larger payments.
Frequently asked questions
What is the maximum OpenAI bug bounty for a critical finding?
An exceptional critical discovery may receive up to $100,000 through the ordinary award and as much as $172,000 when incentives are added.
Which OpenAI products are covered by the new critical-finding range?
It covers core offerings including ChatGPT, Codex and the API Platform. Standard awards for critical flaws can reach $30,000, while higher amounts are available for exceptional discoveries and additional incentives.
Will older unresolved reports receive the new reward range?
Yes. OpenAI will assess reports that have not yet received payment using the revised ranges. It will also cover the gap for awards issued during the 30 days before the announcement if the new calculation would have paid more, including ordinary and loyalty awards.
Why does OpenAI value a reliable proof of concept?
It identifies the flaw, shows how it can be triggered and demonstrates its effects. That gives engineers a practical basis for reproducing the issue and checking a repair.
Does the bounty program replace OpenAI’s internal security testing?
No. It works alongside the Defense Factory, where engineers use OpenAI models to locate vulnerabilities, reproduce them and test corrections. Independent researchers contribute another source of specialist scrutiny.