OpenAI hires outsiders to attack itself: Cathedral joins the next security sprint as the bounty ceiling rises to $172,000
OpenAI announced on November 2 that its next internal security sprint will bring in external Cathedral researchers, and that it is raising the bounty ceiling to about $172,000. The sprint is planned and has not begun; background details rest on third-party reporting.
- 01THE ANNOUNCEMENT — OpenAI published two official blog posts in succession on November 2, announcing the Cathedral sprint partnership and the new bounty policy.
- 02THE CONFIRMATION — Unite.AI and several independent reports checked the bounty figures and multiplier structure and found them consistent.
- 03THE WATCHPOINT — When the next sprint begins, which systems it covers, and whether a retrospective is published.
WHAT HAPPENEDWhat Happened
OpenAI published two official blog posts in succession on November 2: the next internal security sprint will bring in researchers from the external security firm Cathedral to challenge OpenAI's own defences alongside its models and internal teams, finding weaknesses and testing fixes.
The previous sprint is the template: OpenAI committed 25% of its product engineers to hunting weaknesses, using Codex in an isolated environment to reproduce vulnerabilities and prepare patches.
The new bounty policy released the same day raises the median of the reward ranges for medium, high and critical severity by more than four times on average: critical vulnerabilities in the core products ChatGPT, Codex and the API platform pay up to $30,000, exceptional findings remain at $100,000 before tax, and with quality, loyalty and priority-programme multipliers stacked, the ceiling reaches about $172,000.
The new policy is also retroactive: reports not yet paid are calculated directly under the new standard, and reports paid within 30 days before the announcement are topped up if the new standard would pay more.
In the sprint post, OpenAI also reiterated its Daybreak defender programme: $1 billion in subsidised access credits within six months for front-line defenders of critical services.
WHY IT MATTERSThe bounty increase buys an external perspective, not a vulnerability count
OpenAI's logic is in the same sentence in both posts: traditional defence is no longer enough, attackers getting hold of equivalent models is only a matter of time, and defenders must attack their own systems with the strongest models first. The price rise and the external team are two sides of the same strategy.
What is truly valuable is not the $172,000 ceiling but the multiplier structure: a reproducible proof of concept, long-term loyalty, and priority access to new features — only stacked together do they reach the ceiling. OpenAI is buying high-quality reports, not paying by the piece.
WHY IT MATTERSThe sprint is still a plan, and OpenAI's official account omits Cathedral's background
This needs to be stated clearly: this sprint is the planned next round. It has not begun and has produced no vulnerability findings. Reading it as a completed new round of testing would overstate the story's actual progress.
Cathedral's background is also absent from OpenAI's official account: third-party reporting says the company was founded by former Department of Government Efficiency staff and that its agent once infiltrated a third-party business system of Anthropic's within 12 hours. The official blog says only that Cathedral brings a differentiated weakness-finding method; the background rests on third-party reporting.
SIGNAL JUDGMENT — ANALYSISWatch the defence cadence, not the bounty ceiling
Tokenta Signal's judgment: the signal in OpenAI's move is not the $172,000 but the shift in defence cadence from periodic patching to continuous adversarial testing. The stronger the models, the higher the attackers' starting line, and labs can only turn weakness-finding into routine work. The next signal is whether a public retrospective appears once this sprint actually begins.
WHAT REMAINS UNCERTAINWhat Remains Uncertain
EVIDENCEEvidence
Confirmed by two sources, OpenAI's two official blog posts and Unite.AI's report; the original links were opened and verified.
PRIMARY SOURCESPrimary Sources
- OpenAI Official Blog (Security Sprint)
https://openai.com/index/preparing-our-next-security-sprint-to-harden-openai/ - OpenAI Official Blog (Bounty Policy)
https://openai.com/index/rewarding-defenders-with-bigger-bug-bounties/ - Unite.AI Independent Report
https://www.unite.ai/openai-plans-new-security-sprint-and-raises-bug-bounty-rewards/
RELATEDRelated Intelligence
- Microsoft Winds Down Its Oracle AI Cloud Rental as Older Nvidia Chips Fall Behind
- An Unreleased Anthropic Model Proved an 18-Year-Old Quantum Conjecture, Machine-Checked in Lean
- IBM Unveils Quantum System Two in Chicago, Its Most Advanced Computer Yet, Powered by Nighthawk
- Vistra Posts $1.14B Q3 Profit as Power EBITDA Jumps Over 50%, Lifts 2026 Outlook