Anthropic: Threat Actors Abused Claude AI to Extract Secrets from 1.8 Million Android Apps

On September 11, 2026, Anthropic revealed that it had observed multiple threat groups attempting to abuse its Claude AI model for malicious purposes. According to a report by BleepingComputer, these groups included financially motivated actors and state-sponsored espionage teams linked to Russia and China. The attackers reportedly tried to use Claude to extract secrets from roughly 1.8 million Android applications, likely through prompt injection or by leveraging the model's capabilities to assist in reverse engineering app code.

The exact technical methods remain partially undisclosed, but the abuse appears to have targeted sensitive data embedded in Android apps, such as API keys, credentials, or proprietary algorithms. Anthropic stated that it has implemented safeguards and monitoring to detect and block such misuse, but the incident underscores the dual-use nature of advanced AI models.

While the full impact is still under assessment, the potential exposure of secrets from a vast number of apps could have downstream consequences for developers and users, including supply-chain attacks or credential theft. Anthropic has not confirmed whether any data was successfully exfiltrated, and the specific threat actor groups have not been named.

This event is part of a broader trend of cybercriminals and nation-state actors exploring AI tools to enhance their operations. Organizations using AI services should be aware of these risks and ensure they have robust security measures in place.