A prominent technology trade group is urging the U.S. government to revise proposed contract rules for artificial intelligence systems, arguing that the requirements to ensure AI is "truth-seeking" and "ideologically neutral" are not technically feasible and could deter companies from doing business with federal agencies. The Business Software Alliance (BSA), which represents major tech companies including Microsoft, OpenAI, Oracle, Salesforce, and IBM, submitted comments expressing concern that the draft rules, issued by the General Services Administration (GSA), set unachievable standards.
< ]]>The Trump administration's initiative, stemming from an executive order, aims to ensure federal agencies procure AI models that align with principles of ideological neutrality. The GSA is attempting to translate these principles into procurement guidance for government contracts. Under the proposed rules, companies would have to certify that their AI systems meet these standards, undergo government testing, and could face contract termination if found non-compliant.
The BSA contends that current AI technology, particularly large language models, operates on probabilistic responses rather than guaranteed factual outputs, making absolute accuracy and neutrality difficult to ensure. The group stated in its comments that "Current technical realities do not enable guarantees of factual accuracy, as they are inherently probabilistic," adding that terms like "nonpartisan" and "neutral" lack sufficient clarity to be translated into infallible technical measures.
Instead of strict compliance, the BSA recommended a "commercially reasonable efforts" standard, asserting that AI developers should not be held responsible for guaranteeing outcomes beyond their technical control. Furthermore, the alliance pointed out that many companies rely on third-party AI models, making it impractical to enforce federal contract requirements on these external components. The proposed rules, which also address areas like open-source software, intellectual property, data portability, and government testing, could disincentivize AI companies from pursuing federal contracts, potentially hindering the administration's goal of accelerating AI adoption across government services.
Legal experts echo these concerns, with Ryan Frazee, a government contracts partner at Mayer Brown, describing the proposed standards as vague and subjective, noting that bias in AI responses can be "in the eye of the beholder." He also highlighted that the rules appear to misunderstand the probabilistic nature of modern AI systems, deeming the requirement for factually accurate responses unrealistic.
The pushback comes in the wake of a Washington Post analysis that reportedly found left-leaning bias in leading AI chatbots, such as OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8, when responding to contentious issues. This situation also follows a previous dispute between the Trump administration and AI firm Anthropic, where the president labeled the company "radical left, woke" over its restrictions on military applications of its technology. Anthropic had sought to prevent its models from being used for fully autonomous weapons or mass domestic surveillance, leading to a temporary ban on its technology and a subsequent lawsuit from Anthropic alleging retaliatory actions by the government. A federal judge had questioned the administration's justification for the ban, after which Trump reportedly rescinded his national security threat designation.
The GSA has stated it is in the process of establishing acquisition requirements to protect government data within large language model systems and welcomes further stakeholder input on the proposed rule.