OpenAI Prepares New Security Sprint with Cathedral and Quadruples Bug Bounty Rewards
On November 2, OpenAI announced a new internal security sprint conducted alongside the firm Cathedral, whose researchers will use its models to test its defenses, as well as an overhaul of its bug bounty program offering up to $30,000 for critical vulnerabilities, and up to $172,000 with bonuses included.
OpenAI announced two related cybersecurity initiatives on November 2, 2026: a new internal security sprint, this time involving external researchers from the firm Cathedral, and a significant increase in rewards for its bug bounty program, hosted on Bugcrowd. The stated goal: leverage its own models to uncover its flaws before attackers do.
A Sprint with External Researchers
OpenAI explained that Cathedral researchers will use its models alongside its internal teams to “stress-test” its defenses. The company provided no dates, duration, or headcount for the sprint. However, it detailed the previous sprint: 25% of its product engineers were assigned to vulnerability research and remediation, relying on Codex to investigate reports, reproduce flaws in isolated environments, and prepare patches that were reviewed before deployment.
The post justifies this approach by arguing that traditional defenses “are no longer enough” and mentions “recent incidents across the industry,” without naming them. It also noted that OpenAI provides its most capable models to customers of Daybreak, its cybersecurity offering, and pledged $1 billion in subsidized access over six months for defenders of essential services.
Bounties Reaching up to $172,000
On the bug bounty side, OpenAI stated that it is more than quadrupling on average the midpoint of its reward ranges for medium, high, and critical vulnerabilities.
| Severity (OpenAI examples) | Previous bounty cited | New illustrative calculation |
|---|---|---|
| Medium, ChatGPT | $750 | $2,415 |
| High, Prism (non-core asset) | $1,000 | $3,162.50 |
| Critical, ChatGPT (one-click account takeover) | $6,500 | $51,577 |
| Exceptional, ChatGPT (reading other users’ conversations) | $100,000 | $171,925 |
Examples published by OpenAI, presented as illustrative, accessed November 3, 2026.
In practical terms, a critical vulnerability on a core product (ChatGPT, Codex, API platform) now pays up to $30,000 as a baseline. Three multipliers can be stacked: a quality bonus for reports featuring a reliable proof of concept, a five-tier loyalty program, and priority focus campaigns with dedicated reward pools for new features. Pending reports will benefit from the new reward tiers, and bounties paid over the previous 30 days will be topped up. The security team will also attend the PROMPT2PWN event in Oakland on November 20–21.
What to Keep in Mind
The highest figures require stacking every available bonus and remain hypothetical best-case scenarios. The $100,000 cap for an exceptional discovery was already in place; the real shift lies in the “routine” payouts for medium-to-critical vulnerabilities, alongside incentives for well-documented submissions.
Daybreak, OpenAI’s security offering cited in the announcement, made headlines just last week regarding training terms on Daybreak Blue sessions.
Our take
OpenAI’s narrative is consistent: if models are becoming remarkably effective at discovering vulnerabilities, AI labs must first point them at themselves—and pay external researchers well enough to report flaws rather than resell them. Increasing baseline payouts for medium and high-severity bugs, which occur far more frequently than “exceptional” zero-days, is arguably the most tangible improvement for the community. That said, the sprint with Cathedral was announced without a timeline or commitment to publish results, and the most eye-catching sums are merely calculation models rather than actual payouts. We will judge this effort on published disclosures and actual bounties disbursed in the coming months.
Sources
- OpenAI — Preparing our next security sprint to harden OpenAI (November 2, 2026) Primary source
- OpenAI — Rewarding defenders with bigger bug bounties (November 2, 2026) Primary source
- OpenAI on X — security sprint announcement (November 3, 2026) Primary source
- Bugcrowd — OpenAI Security Bug Bounty program Primary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
