OpenAI revised its Security Bug Bounty on 2 November 2026, lifting rewards for Medium, High and Critical findings. The midpoints of those ranges rise by more than fourfold on average, although what a researcher is actually paid still depends on the scope live on Bugcrowd. The same day, the company said researchers from Cathedral Systems would join its next internal security sprint, working with OpenAI's models and its own teams to probe its defences and test fixes.
“should be thinking about how to put these capabilities to work”
Gavin Kliger, co-founder and CEO of Cathedral, to Dealroom · dealroom.co ↗
The headline numbers: a critical bug in ChatGPT, Codex or the API platform can earn up to $30,000, while an exceptional critical finding has a base of up to $100,000 and can reach about $172,000 with bonuses. OpenAI's worked example comes to $171,925, and it describes its sample payouts as illustrative rather than guaranteed. Reports with a reliable proof of concept get a new quality multiplier, a five-tier loyalty scheme (Bronze, Gold and a top tier named 31337) rewards repeat researchers, and time-limited campaigns put bonus pools on newly launched features. Unpaid reports already filed get the new rates, and awards from the past 30 days are topped up.
The sprint is separate from the public programme. The previous one moved a quarter of product engineers, more than 250 people across over 100 service areas, onto finding and fixing weaknesses. It closed 53 urgent or high-priority issues on its first day; 37% of findings were duplicates, false positives ran at 0.81% after dynamic validation, and 0.53% of fixes were rolled back. Codex helped engineers investigate findings, reproduce bugs in isolated environments and draft patches. OpenAI plans to publish the architecture and lessons of this round too, but has not disclosed its start date, length, scope or whether Cathedral is being paid.
Dealroom's story image for OpenAI and Cathedral's security work
Interesting Engineering / Wikimedia Commons · Image: Dealroom
Cathedral was founded by former DOGE staff and, per Reuters as cited by Dealroom, raised $160 million at a $1.4 billion valuation in July, with a16z and Sequoia taking board seats. Its role follows a Washington Post report last week that its AI agents reached parts of Anthropic's sales, HR and IT systems in under 12 hours in an authorised test; Anthropic said Cathedral overstated the national-security angle of that exercise.
About OpenAI
This is not the first time OpenAI has reworked its bounty. It opened a public programme on Bugcrowd in April 2023, a few months after ChatGPT launched, paying about $200 for low-severity findings and up to $20,000 for the most serious. In March 2025 it raised the top reward for exceptional critical findings from $20,000 to $100,000, and it ran a promotion doubling payouts for IDOR reports until the end of April 2025, capped at $13,000 a report. Bugcrowd has hosted the programme from the start, which is why individual payouts still follow the scope it lists at the time.
11
Works credited
27
Models indexed
Models on AIPopDB
OpenAI frames the move as a response to a threat picture in which traditional cyber defence alone is no longer enough, and a spokesperson said the company is among the most attacked targets worldwide. It comes after an August shift of a quarter of production engineers to security infrastructure and a September programme, Daybreak, promising $1bn of model access over six months to outside defenders at 2,000 organisations. BornCity puts the new reward level roughly on a par with Google's general programme, but below the top payouts Apple and Microsoft offer for operating systems.
Sources · 3