The GSA's latest draft regulations on large language models are causing a stir. While improvements are noted, contractors demand clearer definitions.
The General Services Administration (GSA) has rolled out its latest draft regulations targeting large language models (LLMs), and the response has been a mix of cautious optimism and pointed criticism. The draft, unveiled last month, aims to update acquisition rules that were first introduced in March. The primary question on everyone's mind: Will these changes bring the clarity and practicality that stakeholders demand?
Progress Made, But Is It Enough? #
Industry experts and contractors are acknowledging improvements in the new draft. Amy Benson, Vice President at Science Applications International Corporation, highlighted the "meaningful movement" toward aligning rules with commercial best practices. However, as with any regulatory shift, the devil's in the details. Stakeholders argue for further tweaks, specifically around technical definitions and flowdown requirements.
One significant change involves the relaxation of language concerning foreign AI components and the reclassification of roles like LLM Developers and LLM Service Providers. Yet, many contractors feel these adjustments aren't enough. Are these reforms truly matching the pace of technological advancements, or are they merely cosmetic changes?
Clarity and Definitions in the Spotlight #
Megan Petersen from the Information Technology Industry Council voiced concerns over broad data definitions that could lead to confusion in safeguarding government data. She emphasized the need for more precise terms that align with commercial norms. Petersen's call for the exclusion of certain data types, like metadata, from stringent safeguards underscores the industry's demand for practical regulations.
Tim LeMaster from Lookout echoed the need for clarity, urging GSA to define what constitutes "incidental" LLM functionality. Without clear definitions, the risk is that the regulations could stifle innovation by creating uncertainty.
A Call for Practical Examples #
While GSA collects feedback until August 3, experts like David Timm of Burr &. Furman recommend including practical examples within the regulations. The idea is to illustrate what's in and what's out under the new rules. Jessica Tillipman from George Washington University Law School suggests a three-pronged test to identify data requiring additional protections. This approach could provide the clarity and practical guidance that stakeholders crave. The licensing race in Hong Kong is accelerating, but will the GSA catch up with global trends in AI regulation? Or will these drafts remain stuck in bureaucratic ambiguity? As Asia moves first, the U.S. might need to rethink its regulatory playbook if it hopes to lead in AI adoption. Contractors and industry groups are watching closely, eager to see if their feedback will translate into effective policy.
Get AI news in your inbox
Daily digest of what matters in AI.