OpenAI Bug Bounty: What the New $172,000 Top Payout Means

Bug bounties work best when finding a serious flaw and reporting it responsibly is clearly worth a skilled researcher's time. On November 2, 2026, OpenAI raised its stakes considerably. The OpenAI bug bounty now pays critical findings in core products up to $30,000 at base, and exceptional critical findings up to $172,000 with bonuses, alongside new incentives for high-quality reports.

• Average increase: the midpoints of medium, high and critical reward ranges more than quadrupled.
• Critical base: up to $30,000 for ChatGPT, Codex or the API Platform.
• Exceptional base: up to $100,000, reaching $172,000 with bonuses.
• Retroactive: unawarded reports get the new ranges, and bounties from the prior 30 days get topped up where the new ranges pay more.

How the New OpenAI Bug Bounty Payouts Work

The structure is the interesting part. Instead of one flat award, a payout now stacks multipliers on a base bounty. There are three types:

• Quality bonus: rewards reports with a reliable proof of concept that isolates the bug, shows when it can be exploited and demonstrates impact.
• Loyalty bonus: a five-tier program that scales rewards for researchers who report consistently.
• Priority campaign bonus: short-term bonus pools aimed at newly launched features.

OpenAI's published examples make the math concrete. All are illustrative, but they show the scale of the change:

• Medium, ChatGPT: a read-only member uploading files to a Project without write permission moves from $750 to $2,415.
• High, Prism: a flaw exposing a file uploaded to another user's Project moves from $1,000 to $3,162.50.
• Critical, ChatGPT: zero-click account takeover moves from $6,500 to $51,577.
• Exceptional, ChatGPT: viewing arbitrary users' chat messages reaches $171,925 at the top of the stack.

Note that the largest figures assume every multiplier applies, including top-tier loyalty and a priority campaign. A first-time reporter should not expect those numbers on day one.

Why Pay More for Quality Reports?

The quality multiplier solves a real operational problem. Triage is expensive. A clear, reproducible report lets engineers start a fix right away, whereas a vague one eats up analyst hours before anyone even confirms the bug. Paying directly for a solid proof of concept nudges researchers toward the submissions that shorten the time to fix.

That also matters as AI tools make it easier to generate plausible-looking findings at volume. OpenAI's post does not frame the change that way, so this is my analysis, but a bonus for demonstrated impact is a sensible filter for any program facing a rising tide of reports.

How Does This Fit OpenAI's Broader Security Program?

OpenAI describes outside researchers as a complement to its internal Defense Factory work, in which engineers use its models to find weaknesses, reproduce them and test fixes. The logic is straightforward: internal testing and automation cover a lot of ground, and independent researchers catch what they miss by bringing different assumptions.

It is part of a wider trend toward paying more for high-impact findings. Just this week we covered Google's $250,000 Root-in-Cloud reward, and Microsoft's Zero Day Quest shows how live events can focus researcher attention. OpenAI's security team will also be at PROMPT2PWN in Oakland on November 20 and 21, where researchers can test targets in person.

What Should Security Researchers Do Now?

If you already hunt in this space, three actions are worth your time:

• Read the current program scope first. Reward levels depend on whether an asset is core, and rules live on the program page rather than the announcement.
• Invest in your proof of concept. Isolate the bug, show its preconditions and demonstrate its impact, because that is now explicitly worth more.
• Check pending and recent reports. Anything still unawarded gets the new ranges, and awards in the 30 days before the announcement may qualify for a top-up.

For defenders elsewhere, the lesson travels well: tie rewards to the evidence that actually speeds up your fixes. Find more practical analysis in our AI security coverage.

Sources: OpenAI — November 2, 2026; Help Net Security — November 3, 2026.