Big Tech is pushing back against the Trump administration’s effort to stamp out “woke” artificial intelligence — in spite of a recent bombshell report finding strongly left-leaning bias among leading AI chatbots.
A software industry trade group representing Microsoft, OpenAI, Oracle, Salesforce and IBM says the administration is setting AI companies up to meet requirements no one can reliably satisfy — a move they warn could convince some AI vendors that doing business with Washington isn’t worth the headache.
The group, called the Business Software Alliance, is urging the federal General Services Administration to rewrite proposed contract rules that would require AI systems sold to the government to comply with the administration’s “Unbiased AI Principles.”
The trade group said the draft rules, released for public comment last week, are “not feasible to implement,” rely on vague standards and could expose contractors to legal and financial risks.
The move came weeks after an in-depth analysis by the Washington Post found lefty bias among leading AI chatbots on issues from DEI and gay conversion to campaign finance and defunding the police.
News of the BSA filing was first reported by NOTUS.
“Our comments below identify the unintended consequences of the draft clause, which will thwart this important objective and limit federal access to cutting-edge AI products and services,” the BSA wrote in comments submitted July 23.
The fight stems from President Trump’s executive order directing federal agencies to buy AI models that are “truth-seeking” and “ideologically neutral.” The Office of Management and Budget later turned those principles into procurement guidance that GSA is now trying to write into government contracts.
If the rule takes effect, companies selling AI to Uncle Sam would have to certify their systems meet those standards, submit to government testing and risk losing contracts if officials conclude they don’t.
BSA argues that’s easier said than done.
The group says that’s because today’s AI doesn’t work like a calculator. Large language models generate probabilistic responses rather than guaranteed factual answers, making perfect accuracy impossible.
“Current technical realities do not enable guarantees of factual accuracy, as they are inherently probabilistic,” the group wrote. It added that “‘nonpartisan’ and ‘neutral’ are not sufficiently clear guideposts that can be translated into infallible technical measures.”
Instead of requiring compliance, the group urged GSA to adopt a “commercially reasonable efforts” standard, arguing that developers should not be expected to guarantee outcomes that remain outside their technical control.
The filing also says many software companies couldn’t comply with some of the proposed rules because they rely on AI models built by third parties, making it impractical to “flow down” federal contract requirements to companies they don’t control.
The group also warned that rules covering open-source software, intellectual property, data portability and government testing could make AI companies think twice about selling their products to Washington.
BSA said the rules could end up driving AI companies away from federal contracts — undercutting the administration’s own push to speed AI adoption across government.
Ryan Frazee, a government contracts partner at Mayer Brown, told The Post that the proposal asks contractors to meet standards that are difficult to define, much less enforce.
“It’s a very vague and subjective standard. Whether an AI model’s response to a specific question is biased is going to be in the eye of the beholder,” he said.
Frazee added that the proposal also misunderstands how today’s AI systems work.
“They’re not fact retrievers; they’re probabilistic models,” he said.
“The requirement that there needs to be a factually accurate response with that kind of system is unrealistic.”
The Washington Post report recently took leading AI chatbots to task, finding OpenAI’s GPT-5.5 answered nearly every question “exclusively with left-leaning arguments” when asked about 29 hot-button issues. Anthropic’s Claude Opus 4.8 provided just a lefty argument 43% of the time, gave “both-sides” answers 47% of the time — and never served up just a right-leaning answer, according to the analysis.
The proposed new rules came in the wake of Trump’s clash with Anthropic, which he branded a “radical left, woke” company after it resisted giving the Pentagon unrestricted use of its AI systems.
The president ordered federal agencies to stop using Anthropic’s technology, while the Pentagon designated the company a supply-chain risk.
The dispute centered on Anthropic’s effort to keep its models from being used for fully autonomous weapons or mass domestic surveillance.
Anthropic later sued, calling the government’s actions retaliatory, and a federal judge questioned whether the administration had justified the blacklisting.
Trump later said he no longer viewed the company as a national security threat.
“GSA is in the process of establishing acquisition requirements to protect government data in LLM systems, and we welcome further stakeholder input on the rule,” the GSA said in a statement provided to The Post.
