US EditionPublished • loading... • Updated
Experts Say AI-Generated Passwords Pose Security Risks
Irregular found AI-generated passwords have entropy around 27 bits, making them far weaker than they appear and vulnerable to brute-force attacks, researchers said.
- Irregular tested Claude, ChatGPT and Gemini and found highly predictable passwords, a finding Sky News verified, with co-founder Dan Lahav urging, 'You should definitely not do that.'
- Large language models produce patterned outputs because they rely on training data patterns, causing online password-strength checkers to mistakenly rate these predictable passwords as strong.
- Using Shannon entropy and log-probabilities, researchers found 50 prompts to Claude's Opus 4 passwords revealed many duplicates and low entropy, around 27 and 20 bits versus 98 and 120 bits expected.
- Irregular urged developers to review and rotate any passwords generated by LLMs and warned users to change exposed passwords immediately, while Robert Hann recommended passkeys and password managers like 1Password and Bitwarden.
- With AI-generated credentials appearing on GitHub, Irregular found passwords exposed in real projects and Google's NanoBanana showed similar patterns, citing Dario Amodei's warning last year.
Insights by Ground AI
14 Articles
14 Articles
Sky News UKReposted by 
Great Yorkshire Radio
Great Yorkshire RadioAre you using an AI-generated password? It might be time to change it
AI cybersecurity firm Irregular's research, verified by Sky News, found ChatGPT, Claude, and Gemini all produced highly predictable passwords, with repeated and predictable patterns.
·United Kingdom
Read Full Article
Experts warn AI-generated passwords may expose users to security risks
(AA): Cybersecurity experts warned that AI-generated passwords may pose serious security risks, according to an analysis by cybersecurity firm Irregular. ITPro reported on Thursday that the study by Irregular found popular chatbots, including ChatGPT, Claude, and Google Gemini, often produce predictable patterns instead of truly random passwords. Testing revealed repeated outputs and limited variation; one Claude-generated password appeared 18 t…
Coverage Details
Total News Sources14
Leaning Left1Leaning Right0Center4Last UpdatedBias Distribution80% Center
Bias Distribution
- 80% of the sources are Center
80% Center
L 20%
C 80%
Factuality
To view factuality data please Upgrade to Premium
