Research Finds AI-Generated Passwords Are Predictable and Potentially Insecure

New research shared with Sky News by AI cybersecurity firm Irregular warns that passwords generated by popular artificial intelligence models may be far less secure than they appear. The study, verified by Sky News, found that major large language models including ChatGPT, Claude, and Gemini often produce highly predictable passwords that follow repeated patterns. Because these models generate outputs based on patterns in their training data rather than true randomness, the passwords they create may look complex but can be easier for attackers to guess using automated tools.

In testing conducted by Irregular, 50 passwords generated by Anthropic’s Claude resulted in only 23 unique outputs, with some passwords repeated multiple times. Many shared similar structures and prefixes, making them vulnerable to pattern-based attacks. Although common password strength checkers rated these passwords as extremely secure, experts say such tools fail to account for predictable generation patterns.

The issue may extend beyond individual users. Developers increasingly rely on AI coding assistants, and AI-generated passwords have reportedly appeared in code hosted on platforms like GitHub. While many instances involve placeholder or testing code, researchers found cases that could potentially relate to real systems.

Cybersecurity experts recommend avoiding AI tools for password generation and instead using dedicated password managers or built-in random password generators. Others suggest moving away from passwords altogether in favor of passkeys, biometric authentication, or long, memorable passphrases. The researchers are calling on AI companies to integrate truly random password-generation tools into their systems to mitigate the risk.