Artificial intelligence is moving beyond helping bad actors write code or draft messages. Increasingly, it is becoming part of the operational machinery behind cyberattacks, surveillance campaigns, political influence operations, and fraud.
That is Anthropic’s central warning in a new threat intelligence report on malicious activity involving its Claude models between December 2025 and August 2026.
"We’re publishing this work because we believe we have a responsibility to disclose malicious misuse of our services. As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer. The cases range from a network of fake dating apps designed to defraud users to surveillance systems built to identify and monitor dissidents," Anthropic said.
The report documents cases involving suspected state-sponsored groups, financially motivated criminals, commercial surveillance vendors and political operators across seven areas, including cyber operations, surveillance, influence campaigns, scams, biological misuse and conventional weapons development.
AI Is Becoming An Operational Workforce
The most significant shift, according to Anthropic, is that AI can now substitute for portions of the human workforce that previously made sophisticated malicious operations expensive and difficult to execute.
In one case, a single consultant working with Mali’s national security authorities used Claude to help engineer a surveillance platform capable of monitoring communications across the country’s three mobile operators. In Iran, actors used Claude to build surveillance software and a malicious browser extension designed to harvest identities from social networks.
The implications extend beyond governments. Anthropic uncovered a China-based operation that used Claude to power more than 4,700 AI personas across a network of more than 20 dating apps. The fake profiles interacted with at least 25,000 people over two weeks in April, generating roughly 2.36 million messages.
The report also highlights the growing automation of political influence campaigns. Anthropic identified nine influence operations originating from Russia, Iran, Turkey and other regions that targeted audiences across six continents. The campaigns used Claude to create fake social media profiles, build news sites and generate political content, with several operations timed around elections.
In one Malaysia-focused campaign, actors used Claude to build a system that analyzed census and electoral data across all 222 parliamentary constituencies. The operation managed more than 1,000 fake X accounts and used AI to generate fabricated allegations against political figures.
The Economics Of AI-Powered Attacks
Anthropic said the danger is not necessarily that AI is independently deciding whom to target. Humans still make many of the most consequential decisions, including target selection and monetization.
Instead, AI changes the economics of malicious activity. Greater automation can reduce the labor, expertise and cost required to run an operation while allowing attackers to work faster and at greater scale. Anthropic said this could make previously uneconomical targets worth pursuing and encourage higher-volume, lower-touch attacks.
Anthropic said it banned accounts tied to the operations described in the report and strengthened its detection systems, while sharing information with other AI companies and authorities where appropriate.
But the company acknowledged an important limitation: shutting down access to Claude does not necessarily eliminate the underlying threat.
For AI companies, that raises the stakes around safeguards. The challenge is no longer simply preventing a chatbot from answering an obviously dangerous question, but detecting when seemingly ordinary capabilities are being assembled into an industrial-scale operation.
Photo: Shutterstock