Ars Technica AI Security & Safety · Sep 11, 2026 13:02 imp:70 Claude users found ways around safeguards for bioweapons research Some dangerous biology looks much like legitimate research, complicating AI safeguards. Leer el original en Ars Technica AI →