Anthropic: AI agents hit roughly 200 firms' data in one breach

Anthropic's report frames AI as a shift in who gets attacked, not just how. The old logic — go after banks because that's where the money is — assumed costly expertise, so hackers concentrated on high-value targets. Anthropic says AI agents with strong cyber skills are now available at the push of a button, cutting the time and human effort needed to go after less obvious targets. That is why the report concludes many more companies may be attacked.

The supporting cases show how far that work can run. In one intrusion, an attacker with a stolen developer token took full administrative control of a cloud environment in roughly three hours. In another, attackers breached a software provider, extracted more than 2,100 sets of Azure AD authentication tokens across more than 40 corporate cloud environments in about 34 hours, and pulled data belonging to roughly 200 customers. Anthropic also describes a Russian-speaking attacker stealing roughly 26 GB of data and seeking $1.5 million to $2.5 million, and a later campaign targeting roughly 30 AI companies in about four days.

Beyond hacking, the 154-page report covers activity Anthropic disrupted from December 2025 through August 2026, including fraud at scale — one China-based app studio ran more than 4,700 personas and talked with at least 25,000 people over two weeks in April. The stakes look likely to hinge on how quickly defensive tools absorb the same AI leverage, and on whether other security researchers corroborate these patterns.