Anthropic says several hacking groups, including financially motivated criminals and state-sponsored groups linked to Russia and China, have been using its Claude AI model to support malicious activities.
According to the AI company, it recorded different forms of AI misuse between December 2025 and August 2026. These included cyberattacks, influence operations, surveillance, scams, the development of biological and conventional weapons, and attempts to distill AI models.
Anthropic says it disrupted several operations connected to the ShinyHunters hacking collective during the eight months. The group is known for large-scale data theft campaigns that often begin with social engineering and compromised accounts.
One suspected French-speaking ShinyHunters member using the handle “frkoo” reportedly distributed a credential-harvesting system across 10 AWS EC2 workers. The system downloaded Android applications from multiple stores and scanned them for exposed secrets.
Anthropic said the operation downloaded 1.8 million different Android APKs, decompiled them and searched for hardcoded secrets using TruffleHog. Confirmed findings were then sent in real time to a Telegram group organized into more than 100 source categories.
The same actor also used an automated system to collect email addresses belonging to GitHub organizations and used them to obtain GitHub Personal Access Tokens. Anthropic says the credentials collected through these operations were used for most of the confirmed breaches linked to the hacker.
The company also said the actor created a carding website that impersonated French national police and offered stolen payment-card information and an interactive map showing victim addresses.
Other suspected ShinyHunters members reportedly stole AI API keys and used them to breach organizations or conduct reconnaissance. In one case, hackers broke into a software-as-a-service provider and accessed data belonging to around 200 of its customers.
Anthropic says Claude also helped make some attacks much faster. In one case, a suspected ShinyHunters actor spent about 34 hours extracting authentication data and obtained more than 2,100 Azure AD authentication tokens connected to more than 40 Microsoft corporate tenants. Anthropic says AI agents carried out nearly all of the work.
The company also linked Claude-assisted activity to attacks involving a technology provider, an airline and an energy company. In another case, attackers reportedly moved from a single stolen developer token to full administrative access in less than three hours.
Anthropic’s report also describes activity attributed to the Russian espionage group Midnight Blizzard. The group allegedly used Claude to automate malware development, phishing, infrastructure acquisition, persistence, command-and-control operations and data theft.
The group also created an automated feedback process that rebuilt malware when security software detected it. Anthropic says Midnight Blizzard targeted more than 20 government, defense, diplomatic, intelligence and foreign-policy organizations.
A separate Chinese-speaking group, tracked as GTG-10007, reportedly used Claude as an engineering and orchestration layer for a coordinated espionage operation. Its activities included attempts to breach production systems, reconnaissance of government networks across several regions, vulnerability research, exploit development, malware development and the creation of an intelligence-gathering platform.
Anthropic says the group ran autonomous vulnerability-research workflows even while human operators were away. Those workflows reportedly uncovered multiple previously unknown vulnerabilities in a major security product and produced working exploits for several network and security appliance families.
The group targeted around 50 organizations across government, education, retail, energy, technology, healthcare, finance and manufacturing. Anthropic says confirmed compromises included an education technology company, a retailer and a Southeast Asian government agency.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
Anthropic says it disrupted the groups’ use of Claude, banned the accounts involved and strengthened its safeguards based on the observed attacks. The company also said it added new measures to detect similar misuse more quickly and contacted authorities, industry partners and affected victims.
Hackers Used Claude to Steal Data and Find Secrets in 1.8 Million Android Apps
