A Pocket Guide to Mass Destruction: The Tech Giants' Colossal Failure

Security experts reveal that users are bypassing the safety mechanisms of leading language models and extracting instructions for producing biological weapons and lethal toxins, without mandatory federal oversight. The US administration is considering stricter enforcement measures.

Security experts reveal how persistent users manage to bypass the safety mechanisms of leading language models, extracting precise instructions for producing biological weapons and lethal toxins, without any mandatory federal oversight. According to sources familiar with the details, in some cases the chatbots provided highly detailed instructions, ranging from turning viruses into a micro-aerosol with destructive potential to preparing ricin, a toxin banned under international treaties. In some instances, users attached explicit threats to harm others, leading to access being blocked but not always reported to authorities. Experts warn that AI systems are functioning as academic advisors for hostile actors, helping to bridge gaps in experiments that require extensive scientific training. Tech companies face a dilemma: tightening filtering mechanisms could disrupt the work of medical researchers. The US administration has begun considering stricter enforcement measures, and lawmakers have introduced bills requiring mandatory reporting of biological threats.