cathedral systems openai: Ex-Anthropic Bug Bounty Firm Joins Security Sprint

cathedral systems openai is the most awkward security pairing of the week: OpenAI is bringing in Cathedral — the DOGE-alumni startup that previously accessed parts of Anthropic’s systems during a disputed bug-bounty engagement — to help probe OpenAI’s own defenses. OpenAI’s November 2 post says Cathedral researchers will use OpenAI models alongside internal teams in the next security sprint, while the company more than quadruples rewards for critical bug-bounty findings. Reuters, meanwhile, reported that some staffers had warned about operational security lapses for years.

What actually happened

In Preparing our next security sprint to harden OpenAI, OpenAI said it is planning its next internal security sprint with outside partners and that it is partnering with Cathedral. Cathedral’s researchers will use OpenAI’s models with OpenAI teams to challenge defenses. CEO and co-founder Gavin Kliger is quoted praising the collaboration and the idea of using models for continuous defense. OpenAI also said it is improving the Security Bug Bounty program, including more than quadrupling rewards for critical findings.

The same post notes the last sprint dedicated 25% of product engineers to finding and fixing weaknesses, using Codex to investigate findings, reproduce vulnerabilities in isolated environments, and prepare patches. That 25% figure also appears in secondary digests summarizing the plan for a mixed internal-plus-Cathedral next round.

Gizmodo’s November 3 story framed the hire as OpenAI teaming up with a “shadowy” startup founded by former DOGE employees. Cathedral had told the Washington Post it reached sensitive Anthropic information — including Slack messages, job offers, and meeting transcripts — in under twelve hours during a bounty-style test. Anthropic cybersecurity lead Logan Graham replied that the work took “several days,” that model weights and Claude conversations were not affected, and that Cathedral’s approach was “unusual and out of scope” because it relied on multiple third-party vulnerabilities.

Why Reuters’ warning story matters for cathedral systems openai

Reuters’ Deepa Seetharaman reported on November 3 that even before ChatGPT launched in late 2022, some staffers raised concerns about security practices with leaders and the board. Sources described monorepo access patterns, historically wide research documentation on GitHub and Slack, and uneven reviews of internal tools. OpenAI disputed the neglect framing, saying it is heavily targeted and invests in defenses; spokesperson Elana Widmann stressed protecting technology, data, and people as a national-importance responsibility.

The piece also revisited the October Justice Department case of engineer Abdulrahman Alfozan, accused of planting a recording device in an executive conference room and using an internal AI tool to find meeting-related documentation. After Reuters sought comment, OpenAI published the Monday sprint blog — the same cathedral systems openai announcement now sitting on the homepage.

That sequence does not prove Cathedral was hired “because of” Reuters. It does explain why the partnership landed into a news cycle already primed for security skepticism. For readers who follow openai 300 bug bounty and openai safety resigns arcs, this is the hardening chapter with an outside firm that already made Anthropic uncomfortable.

Who Cathedral is, without the mythmaking

Gizmodo names founders Gavin Kliger, Luke Farritor, Marko Elez, and Jack Stein as young DOGE deputies. It cites a reported $1.4 billion valuation after a $160 million round led by Andreessen Horowitz and Sequoia, and notes Reuters’ earlier July reporting that Cathedral’s mission involved strengthening U.S. military cyber capabilities. The company’s public web presence is minimal; its X bio, per Gizmodo, simply says it is “Securing critical American infrastructure.”

For cathedral systems openai coverage, the actionable facts are narrower: OpenAI confirmed the partnership; Cathedral’s Anthropic work is disputed on scope; bounty payouts for critical findings are being raised; and OpenAI says traditional cyber defenses alone are no longer enough.

What this does — and does not — change

A security sprint with outside researchers can surface third-party and identity-path weaknesses product engineers miss. It does not erase industry-wide pressure to ship models fast. Will DePue, a former OpenAI researcher quoted by Reuters, argued leaders “can’t sacrifice velocity for security.” Cathedral’s presence is one bet that AI-assisted offense can widen the defender’s window anyway.

Anthropic’s public pushback also matters for how you read future Cathedral claims. If a firm’s methods are “out of scope” for one lab’s bounty rules, another lab’s engagement letter needs to be explicit about third-party systems, customer data, and model-weight boundaries. OpenAI’s blog does not publish that engagement letter.

What to do as a reader

Read OpenAI’s primary post first, then Gizmodo for the Anthropic dispute, then Reuters for the internal-warning backdrop. If you care about AI-lab security theater versus substance, watch whether OpenAI publishes concrete sprint outcomes the way it shared architecture lessons from the last round.

cathedral systems openai is a process story, not a product launch. No need to invent timelines OpenAI did not give. Cross-link the Anthropic side of the industry rivalry through pieces like anthropic s-1 filing when you want capital-markets context. Nothing here is investment advice — just a sourced map of who is probing whom this week.