Claude users found ways around safeguards for bioweapons research

Some dangerous biology looks much like legitimate research, complicating AI safeguards.

View original article