OpenAI could pay $51,577 instead of $6,500 for a flaw that lets attackers hijack a ChatGPT account

One click, and a user's ChatGPT account falls into a hacker's hands. Whoever discovers such a flaw and reports it to OpenAI could now receive $51,577 instead of $6,500, bonuses included. The company says this is an example, not a sum already paid out. It was one of the illustrations it gave on November 2 when announcing an overhaul of its vulnerability reward program, or bug bounty.

According to OpenAI, the midpoint of the reward ranges for medium-, high- and critical-severity vulnerabilities is more than four times higher on average. A critical flaw in ChatGPT, in Codex (its programming assistant) or in its API platform can now earn up to $30,000 before bonuses.

How high can rewards go?

For an exceptional discovery, the base cap stays at $100,000, already raised from $20,000 in March 2025. Three bonuses are now added on top. The first rewards quality: according to Clubic, a reliable demonstration that isolates the flaw and proves its impact can earn 15% more. The second is a five-tier loyalty program for regular researchers. The third funds priority campaigns on new features. Combined, these bonuses can reach $171,925 in the best case, again as an example. According to Data Studios, OpenAI has not published the amount of each bonus.

The new scale is retroactive. It applies to reports already submitted but not yet rewarded. Researchers who were paid in the 30 days before the announcement will receive a top-up if the new calculation is more favorable to them.

Outside testers armed with OpenAI's models

The same day, OpenAI announced that its next internal security testing campaign would involve researchers from Cathedral Systems, a specialized firm. They will use OpenAI's models alongside its teams to look for weaknesses and verify fixes. In the previous campaign, a quarter of OpenAI's product engineers took part, relying on Codex to reproduce flaws in isolated environments and prepare patches. According to OpenAI, fixes merged into the code had not always reached every production system, so the company kept a human review. The date, duration and scope of the new campaign have not been specified.

Why now?

OpenAI does not present this change as a response to the flood of AI-generated reports, but it stresses reproducible demonstrations, which speed up fixes. The problem is well known: in January 2026, the curl project ended its bounty program, overwhelmed by unreliable reports, often AI-generated. At the same time, AI systems are also finding real flaws. In October, the company XBOW described how it built an attack on Chrome in six days with an OpenAI model.

Companies that use ChatGPT or OpenAI's API have nothing to do for now. Researchers will find the program rules on OpenAI's vulnerability reporting page. It remains to be seen whether these amounts will attract more serious flaws, or just more reports.