Hackers use Claude AI to extract data from 1.8m Android apps
Threat actors automated code analysis to find exposed secrets in days.
15 hours ago
11 September, 20:241 min read
Anthropic has revealed that several hacking groups misused its Claude artificial intelligence model. As reported by BleepingComputer, attackers used the AI model to extract confidential data, secret keys, and authorisation codes from 1.8 million Android applications.
Groups linked to China and Russia
A report by Anthropic's security team notes that both financially motivated hackers and state-sponsored espionage groups linked to Russia and China attempted to abuse the system. By using Claude, the hackers were able to automate application code analysis and quickly discover vulnerabilities.
Artificial intelligence models such as Claude are designed to review software code and assist developers. However, cybercriminals weaponised this functionality, instructing the AI to parse thousands of Android applications to locate sensitive data mistakenly left in open code by developers.
Security measures and response
The report adds that such large-scale analysis of Android apps would typically take weeks or months using traditional methods. By deploying powerful language models like Claude, attackers automated the process and completed it in just a few days.
Following the discovery, Anthropic immediately banned all accounts associated with the attackers and strengthened Claude's safety filters. The company explained that AI security monitoring requires continuous updates, as threat actors periodically find new ways to bypass guardrails.