Anthropic's September 10, 2026 threat-intelligence report describes misuse cases it says it identified and disrupted between December 2025 and August 2026. The company presents these as notable examples, not a representative count of all cybercrime involving AI.
Anthropic says some actors moved beyond asking Claude for isolated advice and used AI-assisted or multi-agent workflows to help run pieces of cyber campaigns. In the cases it described, AI was used for steps including reconnaissance, tool development, phishing infrastructure, and processing stolen information. Anthropic says humans remained involved in choosing targets and reviewing exfiltrated material. These descriptions come from the service provider's own investigation, not an independent census of attacks.
One case Anthropic labels GTG-20006 concerned an actor whose attribution it said was consistent with public reporting on Midnight Blizzard. Anthropic describes AI-assisted workflows that modified and rebuilt malicious tools when security products detected them, and says the actor targeted government, diplomatic, defense and related organizations. That case illustrates a risk: defenders may have less time to respond if an attacker can rework tooling quickly. It does not prove that every attempted operation succeeded or that all attackers now have the same capabilities.
Anthropic said it disrupted the activity it observed, strengthened safeguards, and shared relevant intelligence with authorities and partners where appropriate. The report uses Generative Threat Group, or GTG, as Anthropic's internal label for actors it observed abusing AI; it is not a separate regulatory designation. Organizations should read the case studies as examples of how AI can affect attack workflows, while assessing their own exposure using current security guidance and verified indicators rather than assuming a universal new threat pattern.