The FTC wants ideological bias to be a Section 5 violation, and won't say how it would prove one

In a proposed policy statement released last month, the Federal Trade Commission said it was considering treating ideological bias in AI systems as an "unfair and deceptive practice" under Section 5 of the FTC Act ^[1]. That matters to anyone budgeting AI compliance work, because the same document asserts that this authority supersedes state AI laws and names the Colorado AI Act, which requires risk assessments, transparency disclosures and bias audits before a model ships ^[4][5].

The commission's theory is that consumers expect AI systems to give them information free from bias or ideological manipulation ^[2], and that classifying such bias as unfair or deceptive would let it reach the training data and inputs behind the models ^[3]. What the document does not fully explain is how the FTC would determine that ideological bias exists in a given system ^[6].

That gap is the whole story for practitioners. Colorado's approach, whatever you think of it, tells a compliance team what to produce and when: an assessment, a disclosure, an audit, before release ^[5]. A Section 5 theory produces no artifact list, no filing, and no safe harbour for having done the audit. It produces an open-ended enforcement risk whose evidentiary threshold is unknown even to the agency proposing it ^[6]. Meanwhile the Colorado law does not take effect until 2027, and state lawmakers are already trying to delay or eliminate the audits ^[7]. The plausible near-term outcome is neither regime working: state obligations thinned out, federal standard undefined.

The comment file is not kind. The FTC received more than 300 comments from trade associations, think tanks, individual experts and members of Congress ^[8]. Most criticised the proposal as ill-defined and vulnerable to politically motivated censorship, though some supported stronger rules against bias in AI systems ^[9]. CyberScoop reviewed dozens of critical comments and reported that even ideological allies of the commission raised two objections: that the proposal distracts from real questions about the federal role in regulating AI deception, and that it opens the way to political censorship of model outputs ^[10][11].

Leah Siskind, a former White House digital official and deputy director of the AI Corps at the Department of Homeland Security, now a senior AI fellow at the Foundation for Defense of Democracies, told CyberScoop the statement misses the substantive problem ^[12]. Her research has looked at how authoritarian propaganda is overrepresented in large language model answers, partly because governments deliberately poison the data those systems ingest ^[13]. "There is a really interesting debate here about bias and about accuracy in models and whether that's deceptive or not... but this is not addressing that at all," she said, describing the statement as a power struggle with states and "petty squabbles about which AI model is more woke than the other" ^[14][15]. In her reading, the FTC's role is to police consumer protection violations, not to regulate AI systems, and the proposal stretches Section 5 well beyond its traditional role ^[16].

There is also an asymmetry worth logging. Anthropic, which has clashed with the Trump administration over AI guardrails and military applications, appears more than half a dozen times in the footnotes, often framed as an example of the bias the FTC wants to stamp out ^[17][18]. Elon Musk and xAI are not mentioned at all, despite Musk publicly admitting to intervening when Grok's responses upset him, including on South African race relations and the "MechaHitler" episode ^[19][20]. Grok's only appearance is a footnote quoting an advertisement calling it "your truth-seeking AI companion for unfiltered answers" ^[21]. Zero citations against six or more is a reasonable proxy for who expects a civil investigative demand ^[22].

Watch three things: whether a final statement defines any evidentiary test for ideological bias, whether Colorado's legislature strips the audits before 2027, and which company is named first.