Claude users found ways around safeguards for bioweapons research

Ars Technica2026年9月11日正文已收录本站

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