OpenAI has warned that its future AI models could be so advanced that they could assist individuals with little or no expertise in creating new biological weapons. In a recent blog post, OpenAI acknowledged that, while its models offer positive use cases, such as biomedical research and biodefence, they also present challenges in balancing scientific progress with safeguards against ‘malicious information’, including the potential to assist highly skilled individuals in creating biological weapons. Although physical access to laboratories and sensitive materials remains a barrier, OpenAI cautions that these safeguards are not insurmountable. Johannes Heidecke, OpenAI’s chief security officer, emphasised that the main worry is that advanced AI could enable individuals with limited background knowledge to replicate existing biological threats with which experts are already familiar. To mitigate this risk, OpenAI is emphasising its commitment to taking effective countermeasures. These include stepping up efforts to develop advanced AI models equipped with robust safeguards and controls that can recognise and flag potential threats to human supervisors.
Antonio Cimadoro
