Anthropic's new report details AI-powered disinformation from Russia, Iran and beyond

[JayJ]

Anthropic's September 2026 threat intelligence report, covering activity disrupted between December 2025 and August 2026, details nine separate influence operations built using Claude, originating from Russia, Iran, Turkey, and operators across the Gulf, South Asia, Africa and Europe, targeting audiences on six continents. In one case, a Russian-speaking operator ran a daily pro-Russia, anti-France content pipeline through a Central African Republic radio station tied to the Wagner Group, using Claude to write scripts, draft loyalty-mandated employment contracts for staff, and generate scoring systems to rank which journalists to keep or fire based on how well their content matched the desired political line

A separate French digital advertising agency used Claude to mass-produce roughly 8,900 articles across some 70 fake news websites in about 20 languages, amplified by more than 250 inauthentic social media accounts, shifting political stances depending on whichever client was paying at the time, a pattern the report calls commercial influence-as-a-service. Another operation targeted Malaysia's national elections specifically, using real census and voter data to micro-target roughly 1,000 fake accounts around race, religion and royalty across all 222 parliamentary constituencies, requesting as many as one million artificial views on the sitting prime minister's account

Three separate Iranian state-aligned institutions used Claude to build campaign doctrine, persona systems and target databases tied to what they called a soft war or cognitive warfare program, at one point attributing false claims to Western think tanks including CSIS, Brookings and RAND to make state messaging appear independently corroborated. Anthropic said Claude refused several of the most aggressive requests, including naming real individuals as militants to draw security action against them, though operators sometimes found workarounds by reframing requests in more neutral language. Curious what people think about this specific pattern, does the sheer geographic and political diversity of these operations, targeting six continents across wildly different political systems, suggest this is now a genuinely global and normalized tactic rather than a threat confined to a handful of well-known state actors

Countering misuse of AI: September 2026 / Anthropicwww.anthropic.com