Hackers abused Claude to draw out tricks from 1.8 M Android apps

Anthropic states numerous danger groups, consisting of the economically inspired and state-sponsored espionage groups connected to Russia and China, attempted to abuse its Claude AI design for destructive functions.

The AI business states that in between December 2025 and August 2026, it taped numerous kinds of expert system abuse, consisting of for cyber and affect operations, monitoring, rip-offs, advancement of biological and standard weapons, and design distillation.

Over the eight-month duration, Anthropic interfered with numerous activities connected to the ShinyHunters cumulative, notorious for huge information theft attacks that usually start with social engineering and account compromise.

A supposed French-speaking member of the group that utilized the manage ‘frkoo’ dispersed a credential-harvesting pipeline throughout 10 AWS EC2 employees that downloaded from numerous shops and after that scanned for tricks in 1.8 million Android APKs.

“This pipeline mass-downloaded 1.8 million unique Android APKs from numerous app-store sources, decompiled them, and scanned for hardcoded tricks with TruffleHog,” Anthropic discusses

“Verified findings were routed in genuine time to a Telegram group arranged into over 100 source types.”

The exact same star utilized a different automatic procedure to gather GitHub company e-mail addresses and utilized them to get GitHub Personal Access Tokens (PATs).

The 2 pipelines supplied initial-access qualifications that ‘frkoo’ utilized “for the bulk of the confirmed breaches” related to the hacker.

Anthropic states that ‘frkoo’ likewise established a carding store at policenationale[.]cc that impersonated the French nationwide authorities to offer taken payment-card records, complete cardholder details, and an interactive map of victim addresses.

Suspected ShinyHunters members likewise took AI API secrets and utilized them for breaching other companies or for reconnaissance activity.

In one case, they breached a software-as-a-service company and took information coming from around 200 downstream clients.

Hectic attacks

With the assistance of Claude AI, it took a presumed ShinyHunters danger star about 34 hours to draw out authentication information and get more than 2,100 sets of Azure advertisement authentication tokens connected to over 40 different business Microsoft occupants. According to Anthropic, “AI agents performed nearly all of the work.”

Extra hazardous activity including Claude and credited to ShinyHunters affiliates consists of breaching an innovation company and taking 1TB of information, jeopardizing an airline company, and accessing systems of an energy business.

ShinyHunters moved rapidly after acquiring preliminary gain access to. When it comes to a business software application company, the hackers went to bulk information theft in simply a couple of hours.

In another circumstances, the AI business states that the aggressor moved from a single taken designer token to complete administrative control in less than 3 hours.

Russian and Chinese hackers

Anthropic’s report likewise highlights activity credited to the Russian espionage group “Midnight Blizzard,” which utilized Claude to automate malware advancement, research study, facilities acquisition, phishing, perseverance, command-and-control (C2) operations, and information exfiltration.

The danger star likewise established a feedback loop that reconstructed malware whenever security items discovered it.

Anthropic observed Midnight Blizzard targeting over 20 federal government, defense, diplomatic, intelligence, and foreign-policy entities.

The projects consisted of device-code phishing, ClickFix attacks, DNS pirating through jeopardized hotel Wi-Fi companies, WhatsApp account takeovers, cloud-email theft, and Windows, Android, and iOS malware, with Claude being utilized throughout all attack phases.

Midnight Blizzard automated its operations through AI-driven workflows developed around Claude Code abilities, with the human operator mainly customizing those abilities when they required improvement.

Anthropic likewise explains an espionage operation credited to a Chinese-speaking group tracked as GTG-10007, where Claude was utilized “as the engineering and orchestration layer of a coordinated offensive program involving a variety of tasks,” such as:

• invasion efforts versus production systems
• reconnaissance of foreign-government networks throughout the Middle East, Europe, and Southeast Asia
• a standing vulnerability-research and make use of advancement effort versus significant endpoint-security items
• malware advancement
• constructing an intelligence-collection platform

The GTG-10007 espionage group ran self-governing vulnerability-research workflows while the human operators were away, which discovered numerous formerly unidentified vulnerabilities in a significant security item.

In addition, the automatic effort likewise provided “working exploits for several families of network and security appliances.” The star then leveraged the make use of code versus numerous federal government companies around the world.

The group’s operations targeted around 50 companies throughout federal government, education, retail, energy, innovation, health care, financing, and production, with validated compromises at an education-technology business, a seller, and a Southeast Asian federal government firm.

The AI business keeps in mind that it interfered with the stars’ usage of Claude for damaging activities and prohibited the hazard stars’ account.

Anthropic changed its guardrails based on the observed harmful usage, included procedures to identify future abuse much faster, and got in touch with the authorities, market partners, and victims.

Technology

Construct your security plan for AI-powered attacks

Sign Up With Mikko Hyppönen and security leaders from the NFL, CHANEL, and Atlassian for a two-hour digital top on what AI-speed attacks alter, what protectors ought to stop doing, and how to verify, choose, repair, and re-validate at device speed.

Conserve your seat