Startup says AI agents broke into Anthropic's systems in under 12 hours

A young security company says it broke into the internal systems of Anthropic, the maker of the Claude chatbot, in less than 12 hours. The firm, Cathedral, says it did the work with AI agents. The report comes from the Washington Post, as flagged by Techmeme.

What we know

Cathedral is a months-old cybersecurity startup. It was founded by former DOGE staffers. It says its researchers were able to quickly get into sensitive systems at Anthropic. Cathedral has also told Anthropic what it found.

AI agents are programs that can carry out a series of tasks on their own, instead of answering one question at a time. Here, the claim is that such programs did the probing and the break-in work.

What is not known yet

• Which systems were reached, and what data, if any, was seen.
• How the agents got in, and what weak spots they used.
• Whether Anthropic agrees with Cathedral's account.
• Whether the test was authorized or what rules it followed.

Cathedral's claim is its own. The material available does not include a response from Anthropic or any outside check of the timing.

Why it matters

If AI can find weak spots this fast, any business should expect attackers to use it too. A company with deep security resources was reportedly reached in under a day. Smaller firms often have fewer defenses.

What to do

The basics still help most. Keep software and apps updated so known holes get closed. Use strong, unique passwords, and turn on a second sign-in step where it is offered.