AI Password Generators Found to Produce Predictable Passwords
Cybersecurity researchers found that password-generating features in ChatGPT, Claude, and Gemini produce predictable and insecure passwords despite appearing strong online. Testing showed that AI models generate patterns based on training data rather than truly random sequences, potentially leaving users vulnerable to account breaches.
What happened
Cybersecurity firm Irregular conducted research showing that password generators in AI models including ChatGPT, Claude, and Gemini produce predictable passwords despite appearing strong.
Testing of 50 passwords generated by Claude showed only 23 unique passwords, with one password used ten times. Similar patterns of predictability were found with ChatGPT and Gemini, though to a lesser extent.
AI models generate patterns based on their training data rather than random passwords, and while online password strength checkers rate these passwords as extremely strong, the weakness in patterns is easily recognizable and can be broken relatively quickly even by older computers.
Reported impact
Users who have generated passwords using AI systems are urged to change them immediately, as experts warn the passwords are insecure despite their apparent strength.
Index analysis
The research identified that AI models do not generate random passwords but instead generate patterns based on their training data, creating a false impression of password strength.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.