The Pulse
OpenAI Adds Cathedral to Security Sprint and Raises Bug Bounties
OpenAI plans a security sprint with Cathedral, using its models to test company systems alongside its own teams. The company also raises Security Bug Bounty rewards, with exceptional critical findings on core products eligible for up to $17
AI.info Team · Written with AI · fact-checked ·
“We’re excited to collaborate with OpenAI’s teams, combining their models and security expertise with our research to find weaknesses, test fixes and strengthen this approach to continuous defense.”
Gavin Kliger, CEO & Co-founder, Cathedral
OpenAI plans to bring Cathedral researchers into its next internal security sprint, where they will work alongside company teams and use OpenAI models to test the company’s systems. The announcement, published November 2, also points to a sharp rise in bug bounty payouts: exceptional critical findings on core products may qualify for up to $172,000 when bonuses are included.
The plan combines internal testing with outside researchers and independent reports. OpenAI says the sprint will focus on finding weaknesses and testing defenses, but it does not give a start date, list the systems in scope or report results from the coming work.
Cathedral joins OpenAI’s next security sprint
OpenAI says Cathedral researchers will use its models alongside company teams to challenge the firm’s defenses. The company describes the effort as part of an ongoing push to use AI to find and fix weaknesses, rather than as a one-off review. Cathedral’s work will add external researchers to a process OpenAI says it has already used internally.
For its previous sprint, OpenAI says it assigned 25% of its product engineers to finding and fixing weaknesses. Engineers used Codex to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review. The company did not disclose how many findings the sprint produced, how many patches shipped or how the work changed security outcomes.
Critical bounty awards can reach $172,000
In a separate November 2 announcement, OpenAI said it is raising rewards for its Security Bug Bounty program. The midpoint of reward ranges for medium, high and critical findings will rise by more than four times on average, the company said. A critical vulnerability in a core product such as ChatGPT, Codex or the API Platform can earn up to $30,000; exceptional critical findings on those products remain eligible for up to $100,000 before bonuses, or up to $172,000 with them.
The new incentives include a quality multiplier for reports with a reliable proof of concept that isolates the bug, explains how it can be exploited and demonstrates its impact. OpenAI is also adding a five-tier loyalty bonus and dedicated bonus pools for short-term campaigns tied to newly launched features. Those additions make the payout depend not only on the finding’s severity, but also on the report and the campaign.
OpenAI’s published examples are illustrative, not a promise that a particular finding will receive a stated amount. One example calculates a $51,577 payout for a critical ChatGPT account-takeover report after quality, campaign and loyalty bonuses; another shows $171,925 for an exceptional finding that compromises a security boundary. The company says reports still awaiting awards will use the new ranges, and eligible awards made in the 30 days before the announcement may receive a difference payment.
A larger testing loop, without published outcome data
OpenAI’s approach links model-assisted internal testing with external researchers and a public bounty program. In the prior sprint, the company described a workflow of investigating findings, reproducing them in isolation and preparing patches for review. The new announcement does not say whether Cathedral will follow that same process or how OpenAI will assess the partner’s results.
The bounty changes provide concrete terms for independent researchers, while the planned sprint remains less defined: OpenAI has not published its schedule, testing scope or success measures. The next public milestone for researchers is an in-person opportunity at PROMPT2PWN in Oakland on November 20–21, where OpenAI says its security team will let participants test targets.