Anthropic said it disabled accounts of working scientists who used its Claude AI in ways that could support biological weapons development, part of a broader set of abuses spanning surveillance, scams, conventional weapons design and propaganda. In a report Thursday, the company said the incidents demonstrate model capability but did not assert harmful intent by the individuals involved. Anthropic also removed influence operations tied to Iranian state-linked actors—including one seeking targeting recommendations against U.S. naval forces—and cited Chinese municipal security services using Claude for surveillance and transnational repression. The firm said it has banned implicated users and strengthened safeguards in newer systems such as its Claude Fable 5 model to restrict dual-use biological queries. The disclosures underscore the dual-use risks of frontier AI and are likely to intensify scrutiny from policymakers and regulators.
Related article:
