Before sharing these results publicly, we disclosed the vulnerability to other frontier AI labs via the Frontier Model Forum (
@fmf_org
). Responsible disclosure of jailbreaks is essential, especially as AI models become more capable.

→ View original post on X — @anthropicai