Urgent.News

What's breaking now, across thousands of outlets.

AI

OpenAI backs bipartisan push to address AI biothreats

The company is endorsing three bills to standardize biological data and defend against threats amplified by advanced artificial intelligence.

OpenAI is backing a trio of bipartisan bills aimed at preventing artificial intelligence models from exacerbating threats of biological weapons and synthetic viruses. The company wants to boost access to data necessary for developing AI defenses against both man-made and natural biothreats, and establish guidelines for the secure creation of these tools.

OpenAI exclusively disclosed to POLITICO that the bills receiving new endorsements are the Web of Biological Data Act, the AI-Ready Bio-Data Standards Act, and the SCALE Biology Act. The first two measures have been introduced in both the House and Senate, while the latter originated in the House. This endorsement marks a broader push by OpenAI to legislate on both the risks AI may pose to biological security and the defensive technologies the company could contribute.

Earlier in the day, OpenAI announced its support for a bipartisan House proposal mandating top AI firms to employ independent evaluators to guarantee their models' safety. Richard Johnson, OpenAI's national security risk mitigation lead, likened the company's support for these bills to its stance on AI cybersecurity. He emphasized the need to address safety and risk concerns while simultaneously exploring opportunities for making these tools accessible to defenders.

Recent research has highlighted the potential for AI to aid in the creation of novel pathogens, with a study from Stanford University and the Arc Institute demonstrating the synthesis of 16 entirely new viruses using AI assistance. Another study revealed that Anthropic detected five instances in 2023 of foreign entities abusing its models for dangerous pathogen research, with two of these cases involving bioweapon development.

The endorsed bills seek to expand the National Institute of Standards and Technology's (NIST) role in AI and biosecurity. The SCALE Biology Act would create a new NIST program dedicated to devising measurement standards for biological data and guidelines for responsibly developing biotechnologies and preventing their misuse. The AI-Ready Bio-Data Standards Act would task NIST with crafting a framework ensuring federally funded biodata is compatible with AI researchers' needs.

The Web of Biological Data Act mandates the Energy Department to create a centralized portal for accessing biodata. Johnson highlighted the significance of the data and standards addressed in these bills for advancing techniques to evaluate the implications of powerful AI models on biosecurity. He emphasized the necessity to continuously measure capabilities and risk to compare against industry developments.

Caitlin Frazer, executive director of the National Security Commission on Emerging Biotechnology, a congressionally established entity supporting the bills, underscored the importance of high-quality, AI-ready biological data, likening it to a strategic national resource. OpenAI has previously backed two related bills known as the Biosecurity Modernization and Innovation Act and the Homestake AI Act.

The potential biological threats stemming from AI have fueled calls from the industry to moderate the technology's advancement, with Anthropic CEO Dario Amodei citing the risks as a justification for a more deliberate pace of development in an op-ed over the weekend. Johnson argued that such legislation would facilitate AI research aimed at better understanding the impact of development pace on biosecurity.

"We need to ensure that we're grasping what the capabilities are and safeguard that we don't move too swiftly beyond our capacity to understand how these things could be used and what they could achieve," he stated.

Written by urgent.news from Politico EU's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at politico.com →

More in AI

More from Tuesday 15 September →