Definition
Generative Threat Groups (GTGs) are anthropic’s internal designators for threat actors observed abusing generative AI systems for malicious activity. The September 2026 threat report references GTGs across cyber, fraud, and conventional-weapons cases.
Key Points
- Report measures uplift — AI capability boost via speed, scale, and depth of harm versus without AI
- Cases span December 2025 through August 2026 using Claude Haiku, Sonnet, and Opus (no Mythos-class misuse except one distillation case)
- Notable GTG: gtg-87001 — northern Yemen weapons cell using claude-code for guidance-navigation-control software
- Activity disrupted in each case; safeguards strengthened; intelligence shared with authorities where appropriate