2026-02-12 14:14 UTC
AI agents can detect phishing. They just don’t know not to fall for it.
To address this risk, 1Password built the Security Comprehension and Awareness Measure (SCAM) to test AI models in real-world scenarios. The results:
🚨 Every model committed critical failures
🚨 Some forwarded passwords to attackers
🚨 Others typed real credentials into phishing pages
The good news? A simple 1,200-word security skill dramatically reduced failures.
Now, we’re open-sourcing SCAM. 👉 https://bit.ly/4aocmLC
Replies (4)
-
@dgodon@mastodon.online 2026-02-12 14:37
@1password@1password.social very cool! Even with the skill text added the risks still seem much too great
-
@ThomasHH@chaos.social 2026-02-12 17:07
@1password@1password.social I just love the acronym 😍 Keep up the good work!
-
@jackie@social.linux.pizza 2026-02-12 17:51
@1password@1password.social people are giving LLMs access to their password managers?
-
@Axomamma@mastodon.online 2026-02-12 21:44
@1password@1password.social Burn AI to the fucking ground, please. https://mastodon.online/@tagir_valeev/116057271527521893