Hackers Leverage Claude to Exfiltrate Secrets from 1.8M Android apps

Cybercriminals linked to the ShinyHunters ecosystem used Claude to support a large-scale credential theft operation that downloaded, decompiled, and scanned 1.8 million Android applications for hardcoded secrets.

The campaign shows how AI-assisted workflows can turn exposed mobile-app credentials into a rapid initial-access pipeline for enterprise breaches.

According to Anthropic’s September 2026 threat intelligence report, a French-speaking operator using the aliases MeowSHA, frkoo, and blazespider ran the operation through a fleet of 10 AWS EC2 workers.

The infrastructure collected Android APK files from multiple app-store sources, decompiled them, and scanned the code using TruffleHog, a tool designed to identify exposed credentials, API keys, tokens, and other secrets.

The workflow reportedly identified verified secrets in real time and forwarded the results to Telegram channels organized into more than 100 source categories.

This allowed the operator to sort stolen credentials by type and prioritize keys that could access cloud services, developer platforms, enterprise systems, and other victim resources.

Hackers Exploit Claude to Steal Android Secrets

Anthropic said the Android APK scanning pipeline was paired with a second credential-harvesting stream focused on GitHub. The attackers collected organization email addresses and searched for exposed GitHub Personal Access Tokens. Together, the two pipelines provided the initial credentials used in many confirmed intrusions associated with the operator.

The case highlights a persistent Android security issue: developers sometimes embed sensitive values directly inside mobile applications.

API keys, cloud credentials, backend tokens, analytics secrets, payment-service keys, and private endpoints can be extracted from APK files because applications are distributed to user devices. Obfuscation can slow analysis but does not make a client-side secret safe.

Anthropic described the operation as part of a wider financially motivated campaign involving suspected ShinyHunters affiliates. The group allegedly used AI to accelerate reconnaissance, understand unfamiliar developer and authentication APIs, convert privileged tokens, build bulk-export tools, and collect data across customer environments after compromising SaaS providers.

In one reported incident, attackers used a stolen AI API key for roughly three weeks to support secondary attacks against other organizations.

Anthropic emphasized that the API keys used in these activities were stolen from customer environments, and that Anthropic’s own systems were not compromised.

The attackers’ activity went beyond credential collection. Anthropic observed operators conducting cloud key validation, session replay, token amplification, CI/CD injection, database dumping, and cross-tenant data collection.

In one supply-chain incident, the actors allegedly extracted data from roughly 200 downstream customer organizations. They dumped more than 2,100 Azure AD token sets across over 40 corporate tenants in about 34 hours, with AI agents doing most of the work.

For Android developers, the incident reinforces the need to remove long-lived credentials from client applications. Secrets should be held server-side, protected with short-lived tokens, scoped permissions, application restrictions, usage quotas, continuous monitoring, and immediate rotation when exposure is suspected.

Anthropic said it banned accounts tied to the activity, improved detection measures, and coordinated with authorities, industry partners, and affected victims.

The campaign demonstrates that AI can reduce the time and expertise needed to convert publicly available mobile binaries into a searchable inventory of exploitable enterprise secrets.

Learn 7 Metric-Gated AI SOC Deployment Phases – Download Free AI SOC Deployment Playbook 2026.