Regulating AI "Ideological Bias" Sounds Neutral. It's Not

Key Takeaways

• The FTC released a proposed policy statement that would treat "ideological bias" in AI systems as an unfair and deceptive practice under Section 5 of the FTC Act, potentially allowing the commission to regulate AI training inputs and outputs.
• More than 300 public comments criticized the proposal, with critics across the political spectrum raising two core concerns: that it distracts from genuine AI deception issues, and that it creates a mechanism for politically motivated censorship of AI outputs.
• The statement mentions Anthropic — which has clashed with the Trump administration — more than half a dozen times as an example of ideological bias, while ignoring Elon Musk's admitted interventions in Grok's outputs.
• Legal experts argue the FTC is stretching Section 5 well beyond its consumer protection mandate, and that the proposal conflates subjective questions about AI ideology with the factual deception claims the FTC is actually equipped to adjudicate.

The Federal Trade Commission has proposed treating ideological bias in AI systems as an "unfair and deceptive practice" under Section 5 of the FTC Act, according to reporting by CyberScoop. 

The proposal would allow the commission to regulate the training inputs and algorithmic outputs that power AI systems on the grounds that consumers expect AI to provide information free from ideological manipulation. 

The statement also suggests the FTC believes this authority supersedes state AI laws — specifically mentioning Colorado's AI Act, which requires bias audits before model release.

Over 300 public comments followed. Almost all of them, from organizations across the political spectrum, were critical — not because they disagree that AI bias is a real concern, but because the specific mechanism the FTC is proposing is vague, legally dubious, and almost certainly opens the door to political censorship.

Who Decides What Counts?

The proposal doesn't fully explain how the FTC would determine when ideological bias exists in an AI system. That's not a technical oversight. It's the central problem.

"Ideological bias" isn’t a neutral, measurable quantity. What one political constituency reads as bias, another reads as accuracy. A regulatory framework that allows the government to label AI outputs as unfair or deceptive on ideological grounds is a framework that can be used by any administration to pressure AI companies toward outputs that align with its political preferences.

The selective application of the proposal makes the concern concrete. Anthropic — which has publicly clashed with the Trump administration over AI guardrails and military applications of its technology — appears more than half a dozen times in the document's footnotes as an example of ideological bias the FTC wants to stamp out. 

Elon Musk and xAI are nowhere in the document. Musk has publicly admitted to intervening when Grok's responses upset him, including on topics like South African race relations. Those interventions shaped the model's outputs. The FTC's proposed framework for ideological bias apparently doesn't apply to them.

The Cato Institute's submission identified the core confusion: the FTC is conflating two distinct issues, ideological bias in AI systems and factual deception in marketing. These are not the same thing, and the legal tools appropriate for one don't apply to the other. 

"The FTC is trying to judge AI models' accuracy and performance (two largely subjective variables) in the same way it evaluates dietary supplements' medical-benefit claims or users being charged fees without proper notice or consent," they wrote. "This is an absurd comparison."

A Mechanism With No Safe Hands

The most consistent concern raised across the political spectrum — from conservative think tanks to Democratic members of Congress — is that the FTC is establishing a precedent that would allow any administration, present or future, to reshape AI systems to reflect its political views. That concern isn’t partisan. It's structural.

Government authority to designate AI outputs as ideologically biased and therefore unlawful is government authority over what AI systems are permitted to say. The Trump administration's FTC expressing concern about AI outputs that align with "liberal and progressive values" is one application of that authority. 

A different administration expressing concern about AI outputs that promote nationalism or religious conservatism is another application of the same authority. The mechanism doesn't constrain itself to one political direction. That's what makes it dangerous regardless of who's in power when it's created.

From an internet freedom perspective, the right question isn't whether AI models can be biased. They can. The right question is whether the solution to AI bias is a government agency with the authority to determine what counts as ideological distortion and regulate accordingly. 

That’s not a question whose answer should be yes — not because bias in AI isn't a real problem, but because the cure being proposed concentrates exactly the kind of power over public information that a free internet is supposed to distribute and resist.