Yoshua Bengio @Yoshua_Bengio · 03 Feb 2026

However, safeguards remain imperfect. Attackers can still often find ways to evade them relatively easily. One initiative crowdsourced over 60,000 successful attacks against state-of-the-art models. When given 10 attempts, testers can still generate harmful responses about half https://t.co/oXxXttYgB0

Open on X →