OpenAI’s endorsement marks an expansion of the company’s push for legislation addressing both the risks AI could pose to biological security and the defensive technology it could help create, as public concern about the destructive potential of advanced AI models rises. Earlier Tuesday, the company said it would back a bipartisan House proposal requiring top AI companies to embed outside evaluators to ensure their models are safe. Richard Johnson, OpenAI’s national security risk mitigation lead, said the company’s support for the legislation parallels its thinking around AI cybersecurity.
Recent research has drawn attention to the use of AI to engineer pathogens, including a widely-read study by scientists from Stanford University and the Arc Institute research nonprofit, which demonstrated they were able to design 16 completely new viruses with the help of the technology. Anthropic also disclosed last week that the company had uncovered five cases this year of foreign actors illicitly using its models for dangerous pathogen research that could support bioweapon development. Two of the bills OpenAI is endorsing would expand the National Institute of Standards and Technology’s purview over issues related to AI and biosecurity.
The SCALE Biology Act would establish a new NIST program tasked with creating measurement standards for biological data, and guidelines for securely developing biotechnologies and preventing their misuse. The AI-Ready Bio-Data Standards Act would charge NIST with designing a framework for standards ensuring federally funded biodata is usable for AI researchers. The Web of Biological Data Act orders the Energy Department to establish a centralized resource for accessing biodata.
Johnson told POLITICO that data and standards addressed in the bills are essential for developing better methods of evaluating how increasingly powerful models may impact biosecurity. OpenAI previously endorsed two related bills known as the Biosecurity Modernization and Innovation Act and Homestake AI Act. The potential biological threats from AI have intensified calls from industry to slow advancement of the technology, with Anthropic CEO Dario Amodei citing the risks as a reason to more deliberately pace development in a high-profile essay over the weekend.
Johnson said such legislation would help enable AI research focused on better understanding what the pace of development means for biosecurity.
Extract — continue reading at the source.
