AI Alert: Anthropic Blocks State-Linked Bioweapons & Cyberattack Attempts on Claude

AI developer Anthropic has revealed it stopped state-linked researchers from using its Claude models for potential bioweapons creation and other dangerous activities, including cyberattacks and conventional weapons development. In a fresh threat intelligence report covering December 2025 to August 2026, the company detailed at least five cases of suspected 'gain-of-function' attempts related to biological agents. This groundbreaking disclosure highlights the urgent national security risks posed by powerful AI falling into the wrong hands. The report, published just days after a prominent Anthropic researcher raised alarms about AI's existential risks, underscores the 'dual-use' challenge of cutting-edge AI: tools designed for good can be twisted for harm. Anthropic specifically flagged attempts by groups aligned with Russia, China, and Iran to leverage Claude for everything from engineering more dangerous versions of viruses like chikungunya and orthopoxvirus, often for military-linked projects, to orchestrating sophisticated cyber espionage and propaganda campaigns. The company's 'Responsible Scaling Policy' is meant to prevent such misuse, but the incidents show how determined bad actors are, even trying to bypass safety controls. As AI models like Claude grow more capable, the fight against their misuse becomes a constant cat-and-mouse game. Anthropic has banned the accounts, boosted its safeguards, and shared intelligence with authorities, but experts are now pushing for even stricter controls and international cooperation to manage these 'frontier AI models'. This ongoing struggle highlights a critical turning point for AI governance, as companies and governments worldwide grapple with ensuring beneficial AI development while preventing a new era of global threats.