Anthropic dropped its September 2026 report, revealing it shut down operations where AI was used to coordinate recon, exploitation, and data theft. The company also flagged unauthorized campaigns targeting public Claude models to extract or clone their capabilities.

AI-Powered Attack Campaigns Stopped

According to the report, Anthropic detected scenarios where AI was leveraged to systematically prep attacks—everything from target recon to exploiting vulnerabilities and stealing data. Anthropic says it successfully blocked these operations.

Attempts to Clone Claude’s Capabilities

The company also called out unauthorized efforts to extract or reproduce features from public versions of Claude. These were attempts to access or mimic the model’s functionality outside of its intended use cases.

Midnight Blizzard Campaign Mentioned

The report references a campaign Anthropic linked to the group Microsoft tracks as Midnight Blizzard. The document doesn’t go into further detail on this specific operation.