OpenAI Brings In Cathedral, the Firm Whose Agents Probed Anthropic, for Its Next Security Sprint and Raises Critical Bug Bounties to $30,000
OpenAI said on November 2 that researchers from the security start-up Cathedral will use OpenAI's models alongside its own teams to attack its systems in its next internal security sprint. It also said it is more than quadrupling the midpoint of its bug bounty reward ranges on average, with critical flaws in ChatGPT, Codex or the API now paying up to $30,000 and exceptional ones up to $172,000 with bonuses. Cathedral last month reported reaching Anthropic's internal systems under a bounty, findings Anthropic disputed.
OpenAI has picked, for its own in-house attack exercise, the security firm that publicly clashed with Anthropic last month. In posts dated November 2 and promoted on X early on November 3, the company said it is planning its next internal security sprint and has partnered with Cathedral, whose researchers will use OpenAI's models alongside its own teams "to challenge our defenses." In October, the Washington Post reported that Cathedral had told Anthropic its researchers reached sensitive internal systems under a bug bounty; Anthropic said the access was limited and that Cathedral overstated its findings, as we covered in Cathedral's dispute with Anthropic.
The sprint#
OpenAI says its previous sprint dedicated 25% of its product engineers to finding and fixing weaknesses, using Codex to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review. It says "traditional cyber defenses alone are no longer sufficient," and says it has committed $1 billion in subsidised access over six months for frontline defenders of essential services and digital infrastructure. Cathedral's chief executive, Gavin Kliger, said in OpenAI's post that every company "should be thinking about how to put these capabilities to work."
OpenAI did not say when the sprint starts, how long it runs or what Cathedral is being paid.
The bounty#
In a companion post, OpenAI said it is more than quadrupling the midpoint of its reward ranges for medium, high and critical findings, on average. A critical vulnerability in a core product such as ChatGPT, Codex or the API Platform will now earn up to $30,000. Exceptional critical findings remain eligible for up to $100,000, and now up to $172,000 with bonuses.
The bonuses are new: a multiplier for high-quality reports with a reliable proof of concept, a five-tier loyalty programme, and campaign multipliers for newly launched features. In one of OpenAI's illustrative examples, a one-click ChatGPT account takeover that historically earned $6,500 would now pay $51,577 once all three multipliers apply. Unpaid reports already submitted get the new ranges, and OpenAI says it will pay the difference on bounties awarded in the 30 days before the announcement.
Limits#
- The payout examples are OpenAI's own and described as illustrative; actual awards depend on its triage.
- OpenAI has not published results from its last sprint beyond the share of engineers involved, so we cannot judge how effective the approach was.
- Cathedral's claims about Anthropic remain disputed by Anthropic; nothing in OpenAI's posts addresses them.