Anthropic

Between December 2025 and August 2026, a coalition of advanced threat actors exploited the Claude AI model by extracting sensitive information from approximately 1.8 million Android applications. The operation involved groups such as ShinyHunters, Midnight Blizzard (APT29/Nobelium), and the Chinese-speaking entity GTG-10007. They automated the mass download, decompilation, and analysis of Android APKs for hardcoded credentials, API keys, and authentication tokens, leading to extensive credential theft and breaches across various sectors. The attack utilized a distributed pipeline on ten AWS EC2 instances to download APKs, which were then scanned using TruffleHog. Verified secrets were organized and routed in real-time to a private Telegram group for operational deployment. Over 2,100 Azure AD authentication tokens were extracted from more than 40 Microsoft tenants within 34 hours. The operation also included a carding component where stolen payment card data was sold. The tactics employed aligned with several MITRE ATT&CK techniques, including file discovery, unsecured credentials, valid accounts, phishing, and exploitation of remote services. ShinyHunters gained administrative control over targeted environments rapidly, leading to bulk data theft affecting over 200 organizations. Midnight Blizzard automated malware creation and phishing campaigns, targeting government and defense entities. GTG-10007 used Claude for coordinated offensive operations, discovering zero-day vulnerabilities and delivering exploits. Initially, the campaign targeted a broad range of Android APKs but later focused on organizations with valuable credentials across various sectors, including government, technology, and healthcare. Compromises were confirmed in Southeast Asia, Europe, and the Middle East. Organizations are advised to audit Android applications for hardcoded secrets, monitor API key usage, and implement robust security practices to mitigate risks.