Anthropic says a French-speaking operator suspected of links to the ShinyHunters ecosystem used Claude in an automated operation that downloaded, decompiled and scanned 1.8 million Android APKs for exposed credentials. The campaign moved beyond one-off chatbot queries, integrating AI into a workflow designed to find secrets embedded in mobile app code at scale.
The operation reportedly ran across 10 AWS EC2 instances, collecting APKs from multiple app stores before using TruffleHog to search for keys, tokens and other hard-coded secrets. Validated findings were sent in real time to a Telegram group organized around more than 100 source categories.
Anthropic said the operator also gathered email addresses tied to GitHub organizations and GitHub Personal Access Tokens. Those two collection streams supplied the initial credentials in most of the confirmed compromises associated with the operator, according to the company.
The report points to a sharply compressed timeline for credential-driven intrusions. In a separate operation attributed to the same ShinyHunters ecosystem, attackers extracted more than 2,100 Azure AD token sets from over 40 organizations in roughly 34 hours. Anthropic said AI agents handled nearly all of that work. In another incident, attackers reportedly went from one stolen developer token to full administrative control of a victim’s cloud environment in about three hours.
Anthropic said the malicious operations it analyzed involved the following Claude model families:
- Haiku
- Sonnet
- Opus
The company said it had not identified comparable activity involving Fable or Mythos, apart from a separate case of unauthorized model distillation. Anthropic says it blocked the identified accounts and updated its detection and protection systems.
The report also describes Russian- and Chinese-linked espionage actors using Claude across several stages of cyber operations:
- Reconnaissance
- Phishing
- Malware development
- Vulnerability exploitation
- Data exfiltration
For Android developers and organizations distributing mobile apps, the episode underscores the risk of credentials left in shipped APKs: automated AI-assisted pipelines can now search enormous app collections and turn exposed secrets into potential entry points far faster than traditional manual review.
