The Bad Guy With An AI Named Claude

Anthropic has released an extensive report on misuse of its Claude models that it disrupted between December 2025 and August 2026, covering seven harm areas: cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and distillation.

According to the report, all top Chinese labs made efforts to distill Claude, which Anthropic says it mitigated and stopped. Distillation attempts require large volumes of realistic queries, so DeepSeek, Moonshot and Xiaomi each sent large numbers of user queries directly to Claude; at least Moonshot then returned the results to its users as Kimi outputs. A joint advisory from the NSA, CISA and FBI addressed the same issue two days before the full report. Anthropic also said Zhipu attempted a distillation attack on Claude Fable but switched to targeting Opus after encountering enhanced safeguards.

The report states Claude Haiku, Sonnet and Opus models were used in the cases, and that no malicious activity was found on Claude Fable or Mythos. Named cases include GTG-20006, a Russian espionage operation linked to Midnight Blizzard that used AI to test whether its tools evaded detection and ran automated phishing; GTG-10007, a Chinese-speaking espionage operation that targeted roughly fifty organizations and compromised an education-technology company; GTG-50020, a Russian-speaking financially motivated actor that attacked 30 companies seeking API keys and never obtained access to a pre-release Claude model; and GTG-50027, a consultant in Bamako who used Claude as part of Mali's state intelligence service to target roughly 25 million SIM cards, a system Anthropic said remains deployed. Conventional weapons cases included operations in China, Russia and Yemen.

In a separate matter, Mowshowitz wrote that Trump has now explicitly declared AI existential risk to be a "hoax" on par with the "Russia hoax" or climate change, a statement he described as very bad news and a rhetorical Rubicon.