Artificial intelligence is no longer merely assisting people with tasks. In the wrong hands, it is beginning to run them.
A new report by Anthropic has detailed a striking range of alleged abuses of its Claude models, revealing how state-linked actors, cybercriminals, political operators and researchers are using AI to automate espionage, influence campaigns, surveillance, biological research and even attempts to copy the capabilities of rival AI systems.
AI Turns Cyberattacks Into Automated Operations
One of the most concerning cases involved a suspected Russian state-linked espionage operation targeting Ukrainian and European governments, diplomatic organizations and defense-related entities. Anthropic said the operators used AI across much of the cyberattack chain, allowing adversaries to operate faster and across a broader range of targets with fewer resources.
Iran’s AI-Powered Propaganda Machine
In Iran, Anthropic identified state-aligned influence operations in which Claude was used to build propaganda systems, create multilingual messaging and disguise the origins of state-backed narratives.
The network reportedly transformed government intelligence bulletins into tailored content in several languages and created material designed to appear as though it came from independent voices.
AI Enters the Surveillance State
Anthropic also found AI being used to build and operate surveillance systems.
One consultant working for Malian national security authorities reportedly used Claude to engineer a mass-interception platform capable of monitoring communications across the country’s mobile operators. In China and Iran, AI was also used to process vast amounts of social media data, identify targets and support surveillance operations.
The Bio-weapon Dilemma
Perhaps the most sensitive findings concern biological research.
Anthropic documented five cases involving AI-assisted research with potential dual-use implications, including work involving toxins and pathogens. The company stressed that these cases do not demonstrate an imminent AI-enabled biological threat, but they highlight how difficult it is to distinguish legitimate scientific research from work that could potentially be misused.
China’s AI Race Takes A New Turn
Then comes the battle between AI companies themselves.
Anthropic said operators affiliated with Alibaba carried out what it described as its largest illicit distillation campaign, using thousands of fraudulent accounts to harvest Claude’s reasoning data and help advance Alibaba’s Qwen models.
The campaign peaked at nearly 3 million exchanges a day, with more than 151 million exchanges observed between May and July 2026, according to Anthropic.
AI Is Becoming The Operator
The broader warning from Anthropic is striking: the biggest change may not be AI independently inventing attacks, but its ability to compress the manpower, expertise and time required to conduct sophisticated operations.
Bangladesh Fake News Factory
The report also describes an automated disinformation operation in Bangladesh that generated at least 1,500 headlines, 300 fabricated stories and 1,500 image prompts aimed largely at rural audiences.
The operation promoted the Awami League and attacked political opponents. Anthropic said it found no evidence that the party directed or funded the network. The operator used 29 Claude accounts and software to generate content in batches before distributing videos across Facebook, YouTube and TikTok.
From cyberwarfare and propaganda to surveillance, biological research and AI model development, Claude’s misuse cases show how quickly artificial intelligence is moving from being an assistant to becoming an operational layer for actors pursuing very different, and sometimes dangerous, goals.
