safetyfactbullishOpen models successfully defended against AI-powered cyberattacks and can help defend against millions of cyberattacks dailyClement Delangue02 Aug 2026https://x.com/ClementDelangue/status/2083908468285620415