Anthropic Reports AI-Powered Espionage Operations
Artificial intelligence is rapidly becoming a tool not only for productivity and research, but also for intelligence and surveillance operations.
Anthropic, the company behind Claude AI, says state-backed actors linked to China, Iran and West Africa used its artificial intelligence models to automate parts of surveillance and intelligence-gathering campaigns.
According to Anthropic, the activity was detected and disrupted between January and July 2026. The company said the operations targeted dissidents, ethnic minorities, journalists and diaspora communities, highlighting growing concerns about the use of generative AI in government surveillance and espionage.
The revelations raise an important question: How is AI changing modern intelligence operations?
How Claude AI Was Allegedly Used in Surveillance
Anthropic’s investigation suggests that the misuse went beyond simply asking an AI chatbot to analyse information.
The company said Claude was allegedly incorporated into several stages of intelligence operations, including information collection, social-media analysis, translation, target profiling and personalised communication.
These capabilities could potentially allow intelligence teams to process huge amounts of information much faster than traditional manual methods.
According to reports, targets included pro-democracy figures in Hong Kong, Tibetans, Falun Gong practitioners, Iranian minorities and Iranian political opponents living abroad.
Anthropic said Chinese-linked actors used Claude to analyse social-media activity, assess political sensitivity and identify individuals who could receive additional scrutiny or coercive questioning.
The company also alleged that AI was used to generate personalised messages and translate communications into regional dialects, potentially making recruitment or infiltration campaigns more scalable.
China-Linked Operations Highlight the Scale of AI Espionage
One of the cases described by Anthropic involved actors linked to China.
The company alleged that Claude was used to help track, profile and communicate with potential targets, including Uyghurs and journalists. AI reportedly assisted with drafting messages and translating responses, allowing operators to communicate more efficiently with people in different linguistic environments.
Anthropic also described an intelligence office that allegedly reduced multiple analyst teams to a single office using AI-assisted workflows.
According to the company’s account, the system was capable of producing thousands of investigations every month.
If accurate, such activity demonstrates one of the biggest potential advantages of AI-powered espionage: scale.
Traditional intelligence analysis often requires teams of researchers to manually collect information, translate documents, compare records and create profiles. Generative AI can automate or accelerate many of those repetitive tasks.
Iran-Linked Actors Allegedly Used AI to Profile Thousands
Anthropic also reported an Iran-linked operation involving social-media surveillance.
According to the company’s investigation, Iranian actors allegedly used Claude to develop a malicious Firefox extension designed to collect identities from social-media networks.
Two linked Iranian units reportedly operated 16 accounts, profiling approximately 6,388 Iranians and analysing more than 155,000 tweets to identify opposition-related accounts.
The reported case illustrates another emerging concern surrounding artificial intelligence: AI can potentially connect information from multiple online sources and transform scattered public data into structured intelligence.
For surveillance operations, this could significantly reduce the amount of human labour required to identify patterns, relationships and individuals of interest.
Why AI Could Transform Modern Espionage
The significance of these cases is not simply that governments are using artificial intelligence.
The bigger issue is automation.
AI systems can potentially perform several connected tasks within the same workflow:
- Collect and organise information from public platforms
- Analyse large volumes of social-media posts
- Translate conversations across languages
- Generate personalised messages
- Create structured profiles of individuals
- Identify people according to specific criteria
- Assist with recruitment and infiltration efforts
- Summarise intelligence for human decision-makers
This could allow relatively small teams to conduct operations that previously required considerably more analysts and resources.
In other words, AI may not replace intelligence officers, but it can potentially multiply their capabilities.
The Growing Risk of AI-Powered Surveillance
The development also raises serious concerns about privacy, human rights and digital security.
Social-media platforms already contain enormous amounts of personal information. AI can make that information easier to search, classify and connect.
A human analyst might struggle to examine hundreds of thousands of posts individually. An AI-assisted system, however, can potentially process large datasets much more quickly.
That creates risks for journalists, activists, political dissidents, minority communities and people living abroad.
There is also a broader concern that increasingly capable AI systems could make sophisticated intelligence operations accessible to smaller organisations with fewer resources.
Anthropic Says It Disrupted the Operations
Anthropic said it identified and disrupted the reported activities and closed the accounts associated with the misuse.
The company presented the cases as part of a broader threat-intelligence investigation covering seven categories of AI misuse, including cyber operations, influence campaigns, weapons-related activity, biological research and surveillance.
The report reflects a growing debate within the technology industry over how AI companies should respond when their models are used for harmful or state-backed activities.
As AI capabilities improve, companies face the difficult task of balancing legitimate access to powerful tools with safeguards designed to prevent abuse.
What This Means for the Future of AI and Espionage
The reported Claude cases suggest that the future of espionage may increasingly involve a combination of human intelligence and artificial intelligence.
AI can potentially accelerate information gathering, automate repetitive analysis and help intelligence operators work across languages and enormous datasets.
At the same time, the technology introduces new challenges for privacy, cybersecurity and human rights.
The cases described by Anthropic should therefore be viewed as part of a much larger trend: artificial intelligence is becoming an increasingly important component of modern intelligence operations.
However, an important caveat remains. The reported operations are allegations from Anthropic’s internal threat-intelligence investigation, rather than independent judicial findings. The company says it disrupted the activities after detecting the misuse.
As governments, technology companies and security researchers continue to confront AI-related threats, one thing is becoming increasingly clear: the same technology capable of transforming business, science and communication can also dramatically increase the scale and speed of surveillance and espionage.
The global race over AI is therefore no longer just about who develops the most powerful models. It is also about who controls how those models are used and how effectively misuse can be prevented.
