A new report from Anthropic, dated September 10, 2026, details how AI is being leveraged for large-scale cyberattacks, fraud, and data theft. While the methods remain traditional, the economics of these attacks have shifted: tasks that previously required teams of people and weeks of work can now be executed by a single individual using AI tools in a matter of hours.
In one case, a network of over 20 Chinese dating apps used AI personas to engage users and extract money. The system automatically generated responses, managed matches, and simulated activity, while human workers were hired only for tasks like video calls. Revenue was diverted through hidden browsers that bypassed App Store and Google Play moderation.
Other findings include a hacker who scanned 1.8 million Android app files to find hardcoded API keys, leading to the compromise of 40 corporate clouds. Another operator targeted AI developers to access unreleased models, while others created fake accounts on crypto exchanges by intercepting KYC processes. Some AI labs were also found using Claude via proxy to power their own services.
The report highlights a campaign targeting European political organizations and media outlets, where a single operator used a custom Rust-based scanner to find access keys. This resulted in the theft of 140,000 voter records and the creation of a massive de-anonymization database. The attacker operated entirely using stolen resources, incurring no personal costs.
Anthropic notes that while their models have safety guardrails, attackers bypass them by breaking malicious requests into smaller, harmless steps. The company also observed that the complexity of corporate infrastructure no longer serves as an effective defense, as AI agents can now map and exploit unfamiliar environments in just a few hours.
