Our latest model has also made improvements on safety. It hallucinates less and is more resistant to jailbreaks, as shown in our automated red-teaming evaluation.

→ View original post on X — @anthropicai