AI-Powered Hackers Exploit Claude to Harvest Credentials from Nearly 2 Million Android Apps

Cybercriminals linked to the ShinyHunters underground market have employed the artificial‑intelligence model Claude to automate a massive credential‑theft operation that targeted roughly 1.8 million Android applications. By combining the language model’s code‑analysis capabilities with a custom workflow, the group was able to locate and extract hard‑coded secrets embedded in the apps at an unprecedented scale.

The attackers began by mass‑downloading publicly available Android packages from various app stores and third‑party repositories. Each binary was then decompiled into readable source code, a step that traditionally requires considerable manual effort. Automated scripts fed the decompiled files to Claude, which parsed the code, flagged strings that matched patterns typical of API keys, authentication tokens, and other sensitive credentials.

Claude’s role went beyond simple pattern matching. The model was prompted to rank discovered secrets by likelihood of value, suggest possible target services, and even generate scripts that could be used to validate the harvested tokens. This AI‑assisted triage allowed the operators to focus on the most lucrative finds, dramatically reducing the time needed to turn raw code into actionable data.

Hard‑coded secrets have long been a weak point in mobile app security. Developers sometimes embed API keys or service credentials directly in source files for convenience, inadvertently exposing them when the app is distributed. Past incidents have shown that such leaks can enable attackers to impersonate legitimate services, exfiltrate user data, or abuse cloud resources. The ShinyHunters campaign underscores how the sheer volume of vulnerable apps can be weaponized when paired with sophisticated AI tools.

Security experts say the operation illustrates a broader trend: artificial‑intelligence models are increasingly being co‑opted for illicit purposes. While Claude was designed to aid developers, its ability to understand and manipulate code makes it a potent asset for threat actors seeking to automate reconnaissance and exploitation. The incident has prompted calls for stricter secret‑management practices, such as using environment variables, secure vaults, and automated scanning of code before release.

Researchers continue to monitor the ShinyHunters ecosystem for further AI‑driven campaigns, and law‑enforcement agencies are reportedly investigating the network behind the attacks. In the meantime, industry groups are urging mobile developers to adopt tooling that can detect embedded credentials early in the development pipeline, and to consider the security implications of integrating AI assistants into their workflows.