The new ceiling, and the new floor
OpenAI raised the rewards in its Security Bug Bounty program on 2 November, saying it is “more than quadrupling the midpoint of our reward ranges for medium, high, and critical findings on average.” A critical vulnerability in a core product — ChatGPT, Codex or the API Platform — will now earn up to $30,000. Exceptional critical findings on those targets stay eligible for up to $100,000, and up to $172,000 once bonuses are applied.
The bonuses are the structural part. Three multipliers stack on top of the base award. A quality multiplier rewards reports with a reliable proof of concept that isolates the bug and demonstrates its impact. A five-tier loyalty programme pays scaling bonuses to researchers who report repeatedly. Priority campaigns carry dedicated bonus pools attached to newly launched features.
What that looks like in practice
OpenAI published four worked examples against historical awards, which it describes as illustrative. A medium-severity finding in ChatGPT — a read-only project member able to upload files without write permission — was historically worth $750; under the new scheme the company puts it at $2,000 with a 1.15 quality bonus and a 1.05 bronze loyalty tier, for $2,415.
The critical example is the sharper one. A one-click ChatGPT account takeover was historically awarded $6,500. OpenAI’s new calculation is $30,000 multiplied by a 1.15 quality bonus, a 1.30 priority campaign bonus and a 1.15 top-tier loyalty bonus, which comes to $51,577 — close to eight times the old figure. At the exceptional tier, the ability to view arbitrary users’ chat messages is shown as $100,000 with the same multipliers, reaching $171,925.
Retroactive, and a date in the diary
The change reaches backwards as well as forwards. Reports already submitted but not yet awarded will be assessed under the new ranges, and for bounties paid in the 30 days before the announcement OpenAI says it will pay the difference where the finding would have earned more under the new base and loyalty awards.
The company frames the programme as complementary to its internal Defense Factory work, where its own engineers use OpenAI models to find weaknesses, reproduce them and test fixes. It also says its security team will be at PROMPT2PWN in Oakland on 20 and 21 November, where researchers can test targets in person.
What to watch
A bounty schedule is a statement about where a company thinks its own testing falls short, and the categories that got the biggest increases are the useful signal here. The measurable follow-up is whether the higher ceiling changes what gets reported — more critical findings in ChatGPT, Codex and the API Platform would suggest the price was the binding constraint.
