Anthropic, the company behind the Claude AI assistant, has found that two out of every three accounts it bans are tied to AI-driven cyberattacks. The figure—67% of all banned accounts—underscores how quickly generative AI is being repurposed for hacking, phishing, and other malicious activity.
What the data shows
The company tracks account bans on its platform to enforce usage policies. Its latest internal data reveals a clear trend: the majority of accounts that get the boot are not casual rule-breakers but active cybercriminals using AI models to automate attacks. The attacks include generating convincing phishing emails, writing exploit code, and scaling up credential-stuffing attempts. Anthropic did not name specific organizations or individuals behind the accounts.
How Anthropic responds
Anthropic uses automated systems and human reviewers to spot suspicious behavior. Once a flagged account is confirmed to be involved in cyberattacks, it is permanently banned. The company also feeds threat intelligence back into its safety systems to catch similar patterns earlier. The 67% number comes from a review of all bans over a recent period—the exact timeframe has not been disclosed.
The finding adds concrete numbers to a worry that has shadowed the AI industry since ChatGPT launched. Policymakers have warned for months that AI lowers the barrier for cybercrime. Anthropic’s data suggests the fear is justified, at least on its own platform. Other AI companies, including OpenAI and Google, have reported similar but less specific concerns about misuse.
What’s next
Anthropic says it is continuously updating its detection methods and working with cybersecurity partners. It has not said whether it will publish regular transparency reports on banned accounts or share more detailed breakdowns of attack types. For now, the 67% figure stands as a quiet reminder that the AI-powered offensive is already here.
