OpenAI Executive Cautions About Rising Threat of AI in Biological Weapons Development

An OpenAI executive has cautioned that the next generation of its large language models might facilitate the creation of deadly bioweapons by individuals with minimal scientific expertise. Johannes Heidecke, Head of Safety Systems, indicated in an Axios interview that upcoming models could receive a “high-risk classification” due to their advanced capabilities. OpenAI is enhancing safety measures to prevent misuse, acknowledging the concern that its models might enable those with limited knowledge to develop lethal weapons. While the company is not apprehensive about entirely new weapon types, it worries about replicating familiar threats. A key challenge is balancing the potential for groundbreaking medical advances with the risk of harm. OpenAI emphasizes the need for nearly flawless testing protocols to evaluate new models before public release. Similarly, rival Anthropic has heightened safety measures for its AI, recognizing the growing risks associated with increasingly powerful models.

Source link