Anthropic claims that Claude was deployed in cyber-espionage campaigns targeting more than twenty organizations, as well as to design a surveillance system in Mali covering about 25 million SIM cards.
Key points
Anthropic says a Russian-speaking operator used Claude-piloted workflows to carry out espionage operations against more than 20 organizations.
Sinophone operators used Claude for automated vulnerability research, identifying more than a dozen potential "zero-day" flaws in a month.
A Malian consultant relied on Claude to build a surveillance infrastructure covering around 25 million SIM cards.
Cyberattacks involving Claude
In its September threat intelligence report, Anthropic explains that a Russian-speaking operator using the pseudonym “JackPoterz” used custom AI workflows to automate a large part of an espionage campaign. The campaign targeted Ukraine and Europe.
The targets: ministries, defense and intelligence agencies, embassies and diplomatic missions, think tanks, and defense groups, with spillovers into other regions. Claude was involved at multiple points along the attack chain.
Anthropic also identified Chinese-speaking operators who used Claude as an engineering and orchestration layer for offensive operations, including vulnerability research against security products and network equipment.
A continuously running workflow produced on its own more than a dozen potential zero-day vulnerabilities in a month, while parallel systems handled reconnaissance, malware development, and intelligence collection.
Also to read: OpenAI’s two-week freeze ended, and then Altman discussed alignment with the pace of rivals
Anthropic warning
According to Anthropic, these cases illustrate how AI lowers the level of expertise and resources needed to carry out sophisticated attacks, enabling individuals or small groups to conduct operations that were previously reserved for much larger teams. Some intrusions would have required only two to three hours.
A second dossier focuses on internal surveillance in Mali. Anthropic says that an independent consultant, likely based in Bamako and working with the state intelligence agency ANSE, used Claude as the primary engineering “workforce” for “Lakana 360,” a platform monitoring about 25 million SIM cards.
The platform ran on local infrastructure and models. Claude provided the software architecture and development support for a system that, according to Anthropic, could collect call histories, SMS messages, and voice streams, while generating intelligence dossiers without requiring the presentation of a judicial warrant.
Anthropic shut down the account in question and deployed new detection mechanisms to prevent similar uses. However, the company specifies that these measures did not neutralize the surveillance platform already installed on site. The September report builds on earlier intelligence publications from Anthropic, in March, August, and November 2025, documenting a shift from chatbot-style assistance to increasingly automated cyber operations.
More to follow: Apple’s iPhone Duo takes on two risks of early demand at $1,999