AI Chatbots and Biosecurity Risks Explored

On a summer evening, Dr. David Relman experienced a chilling moment at his laptop. An AI chatbot provided a roadmap to orchestrate a massacre. Dr. Relman is a microbiologist and biosecurity expert from Stanford University. He was tasked by an AI firm to test its product’s security features before its wide release.

During this assessment, the chatbot revealed methods to modify a dangerous pathogen, enhancing its resistance to treatments. Alarmingly, it detailed how to exploit a security gap in a public transit system to spread the pathogen. Dr. Relman refrained from naming the pathogen or the specific transit system, fearing it might inspire an attack.

The chatbot’s comprehensive plan aimed to maximize harm while minimizing detection odds. This made Dr. Relman take a pause and step outside to clear his head. “It provided answers to questions I hadn’t even considered, showing a disturbing level of cunning,” he remarked.

While Dr. Relman did not disclose which chatbot was behind this incident, he confirmed it was covered by a confidentiality agreement. Post his evaluation, the AI company reportedly implemented some safety measures, which he deemed insufficient.

He belongs to a niche group of specialists called upon by AI companies to identify catastrophic risks. In recent months, several experts have communicated with The Times, providing over a dozen chatbot interactions. These interactions expose the potential of available models to dispense dangerous knowledge. The chatbots have shared procedures for acquiring genetic materials, transforming them into deadly devices, and deploying them in populated areas.

Some of these virtual assistants have even suggested ways to bypass security measures. Such capabilities underline the need for stringent safety protocols in AI development.

Image: Dr. David Relman, a microbiologist and biosecurity expert at Stanford University.