A security firm said an AI model did most of the work in a water utility break-in
Dragos reported in May 2026 that attackers used Claude to build a 17,000-line attack framework.
Kivorent News desk, published , 1 min read
17,000lines in the attack framework, per Dragos
This is a look back at an event from . We wrote it on November 11, 2026, from the sources listed at the end.
The industrial security firm Dragos disclosed on May 8, 2026, that attackers who broke into a water utility used Anthropic's Claude as their main technical tool. The model built an attack framework of 17,000 lines of code, according to AI Weekly's quarterly recap.
The summary we read did not include a response from Anthropic.
Why it mattered
Security researchers had warned that AI models would lower the skill needed to attack critical infrastructure. This was a documented case involving a real utility.
It added urgency to two things that followed. Google released its next security-capable model to defenders first, and the Justice Department set up a task force that names attacks on critical infrastructure as a priority.
Sources
- The state of AI, Q2 2026, AI Weekly
AnthropicInternet, outages and securityAI agents
Spotted an error? Tell us and we will fix it.