Overview
- The company published a detailed misuse report Thursday that says its threat‑intelligence team identified dozens of suspicious efforts from December 2025 through August 2026 and disrupted multiple operations seeking help from Claude for cyberattacks, surveillance, influence campaigns and biological research.
- Anthropic says it barred the specific accounts, worked with partners to take down relay networks that routed around regional blocks, and shared findings with other labs and government authorities to stop ongoing abuse.
- The report includes five biological case studies, one involving a gain‑of‑function grant application about the chikungunya virus, and describes other probes into bird flu, orthopoxviruses and toxins that the company judged as dual‑use and potentially dangerous.
- Anthropic says its newest Claude models, including Fable and Mythos classes, were not used in the cited cases except for one illicit distillation attempt, and the company says it has applied stricter restrictions to recent models to limit dual‑use biological queries.
- The disclosure follows a high‑profile researcher resignation warning about runaway AI and has intensified calls from lawmakers and investigators for mandatory reporting, tighter rules and industry coordination even as Anthropic prepares a confidential IPO and expands its underwriting syndicate.