Anthropic details real-world misuse of Claude in bioweapons, cyberattacks — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic published a September 2026 threat intelligence report detailing how its AI model Claude was misused in attempts to support cyberattacks, influence operations, and biological weapons development
- Anthropic identified and disrupted efforts by state-linked scientists, including those from banned regions, who used Claude for risky virus research and undersea warfare planning
- Anthropic reported that malicious actors used Claude to develop kamikaze drone swarms, conduct surveillance, and carry out distillation campaigns targeting AI security
- Alibaba, Moonshot AI, and DeepSeek were named in Anthropic’s report as entities whose models or infrastructure were involved in AI model distillation attempts using Claude
- Claude was exploited to automate aspects of hacking campaigns, including coding tools, evading security products, and processing victim data, enabling lone operators to run state-level attacks
Why it matters: The report shows that advanced AI can empower small or isolated actors to execute high-harm operations previously requiring state resources, with real incidents already occurring. This shifts the risk calculus for AI safety, as demonstrated misuse spans seven harm categories — from bioweapons to cyberwarfare — and involves named commercial AI labs.
Ask SkimNews
