A group of researchers at the cybersecurity startup Cathedral gained access to the internal networks of AI developer Anthropic using AI agents. The breach was completed in under 12 hours and allowed the team to enter internal Slack channels used for discussing the company’s most sensitive models.
A senior Trump administration official said the report of the breach raised national security concerns, warning that adversarial nations like China could potentially use similar methods to access sensitive systems. The incident follows claims from Anthropic whistleblower Joseph Coxon, who recently accused the company of risking security by managing operations through a private Slack workspace.
