OpenAI strengthens its defenses by joining forces with cathedral systems, a young cybersecurity company founded in particular by former members of the American DOGE. The latter will participate in the group’s next internal security sprint to actively search for vulnerabilities in its infrastructures and products.
Recently, Cathedral attracted attention after successfully breaching some of Anthropic’s internal systems as part of a breach program. The researchers gained access to multiple business, HR, and IT environments through third-party software. Anthropic clarified, however, that Claude’s conversations, source code, and the weights of his models had not been compromised.
Artificial intelligence used to attack defenses
At OpenAI, Cathedral will directly use the company’s models to test its protections. “Their researchers will use OpenAI models together with our teams to test our defenses”explains the group.
This approach is part of a strategy already underway: OpenAI bought Promptfoo to strengthen the security of its AI agents and mobilizes its own models to detect vulnerabilities before their exploitation. The issue has become even more sensitive as OpenAI agents have recently weakened certain tracking mechanisms during internal testing. Finally, as a reminder, OpenAI had already had to fix a bug affecting ChatGPT and Codex on Mac.
Bonds that could exceed $170,000
OpenAI is also taking advantage of this crackdown to strengthen its bug bounty program: average bounties for medium, high, and critical bugs have now more than quadrupled. A rare vulnerability affecting ChatGPT, Codex or API can now generate up to $100,000, or even around $172,000 with the various bonuses.
With Cathedral, the company led by Sam Altman takes an increasingly offensive approach to cybersecurity and surrounds itself with big names in the sector. Will this be enough to prevent breaches and cyberattacks? Given the increasing ingenuity of hackers, nothing is less secure…
