Anthropic's new threat report: AI let a criminal group exfiltrate 2,100 corporate access tokens in 34 hours

Anthropic published its latest threat intelligence report on September 10, 2026, detailing Claude misuse cases it detected and disrupted between December 2025 and August 2026 across seven harm categories. The most severe involved a Russia-linked campaign (internally labeled GTG-20006) that hit more than 20 government and defense organizations, exfiltrating over 300,000 national identity records and more than 500,000 commercial registry records. A separate criminal group affiliated with ShinyHunters (GTG-50014) downloaded 1.8 million distinct Android APKs and dumped more than 2,100 Azure access-token sets across at least 40 corporate tenants -- in roughly 34 hours.

The skill-gap argument, with numbers attached

Anthropic's own framing is direct: "everything connected to the internet is a potential target for exploitation." What the report actually documents is less about new attack techniques and more about speed and reach -- a 34-hour window to compromise 40 corporate tenants' worth of credentials is a timeline that used to require a team with real operational infrastructure, not a single actor with access to a capable model.

Not just intrusion -- influence operations too

A third case (GTG-54002) used AI to run an influence operation that published more than 8,900 articles across roughly 70 fabricated news sites in about 20 languages. Taken together with the intrusion cases, the report is Anthropic documenting, in its own numbers, the two things enterprise security and risk teams are already worried about in the abstract -- credential-theft campaigns and disinformation production -- actually happening at the scale and speed AI enables, not staying hypothetical.