Summary
- The FTC is considering classifying ideological bias in AI systems as an ‘unfair and deceptive practice’ under Section 5 of the FTC Act.
- The proposal also asserts federal regulatory authority over state AI laws, including Colorado’s AI Act which mandates risk assessments and bias audits before model release.
- Critics from both left and right argue the statement is poorly defined, legally overreaching, and risks enabling politically motivated censorship of AI outputs.
- Observers note the statement disproportionately cites Anthropic as an example of ideological bias while omitting documented interventions by Elon Musk in Grok’s outputs.
- The most widely shared concern across commenters is that the precedent could allow any administration to reshape AI systems to reflect its political preferences.
What the FTC is proposing
In a proposed policy statement released last month, the Federal Trade Commission signalled it is considering treating ideological bias in AI systems as an unfair and deceptive practice under Section 5 of the FTC Act. The argument is that consumers have a reasonable expectation that AI systems will deliver information free from bias or ideological manipulation. If formalised, the policy would potentially give the commission authority to regulate the training data and algorithmic inputs that shape AI model behaviour. Notably, the statement does not fully explain how the FTC would determine when ideological bias is present in a given system.
Federal preemption of state regulation
The proposal goes further than bias regulation alone. The FTC statement suggests its authority supersedes existing state AI laws, singling out Colorado’s AI Act — legislation that would require risk assessments, transparency disclosures and bias audits before model release, taking effect in 2027. State lawmakers in Colorado are already seeking to delay or remove the audit requirements before the law comes into force. The FTC’s position on preemption adds another layer of complexity to an already uncertain regulatory landscape for organisations deploying AI.
A broad coalition of critics
The FTC received more than 300 public comments on the proposal. While a small number of commenters supported stronger rules against AI bias, the majority criticised the statement as ill-defined and prone to politically motivated misuse. Two concerns emerged consistently across the ideological spectrum: first, that the proposal sidesteps substantive questions about AI deception and accuracy; and second, that it could enable government censorship of AI outputs under the guise of consumer protection.
Selective framing raises questions
Several observers highlighted an apparent inconsistency in the statement’s sourcing. Anthropic — which has had friction with the Trump administration over AI guardrails and military applications — appears in footnotes more than half a dozen times, frequently as an example of the ideological bias the FTC seeks to address. By contrast, xAI’s Grok model is mentioned only in a footnote quoting a product advertisement. This is despite publicly documented instances of owner Elon Musk intervening in Grok’s outputs on specific topics, including South African race relations and the use of certain offensive terms, to align responses with his personal views. Neither Musk nor xAI is named elsewhere in the document.
The legal and conceptual weaknesses
Leah Siskind, a former White House digital official and senior AI fellow at the Foundation for Defense of Democracies, told CyberScoop that legitimate questions exist about AI companies’ obligations to consumers — including whether models must provide accurate information and resist deliberate data poisoning by authoritarian actors. But she characterised the FTC statement as failing to address those questions, describing it instead as a jurisdictional power struggle with states and a focus on ‘petty squabbles about which AI model is more woke than the other.’ She added that the statement appears to stretch Section 5 well beyond its traditional consumer protection role to compensate for a lack of congressional AI regulation. The International Center for Law and Economics noted the statement offers little practical guidance on how deception authority would apply to AI, and raises unresolved questions about when AI providers may be exercising First Amendment-protected speech. The Cato Institute’s submission argued the FTC is conflating two separate issues — ideological bias in AI outputs and factual deception in marketing — calling the comparison to cases involving undisclosed fees or false medical claims ‘absurd.’
Where bipartisan agreement ends
The conservative America First Legal Foundation argued the proposal should be adopted in full, asserting that frontier models have been programmed to favour progressive values. However, the R Street Foundation, also a conservative-leaning group, pushed back, noting the consumer expectations rationale is typically applied to omissions of material information, and that most large language models already carry disclosure statements making it difficult to establish consumer deception. Bipartisan congressional concern was raised separately: Representatives Gottheimer and Lawler urged the FTC to ensure that civil rights-related bias mitigation — preventing models from discriminating by race, religion, gender or age — remains permissible under any final framework.
Why it matters
For CISOs, the immediate relevance is regulatory uncertainty rather than a near-term compliance obligation. If the FTC formalises this policy, enterprise AI deployments could face scrutiny not just over data security and accuracy but over the ideological character of model outputs — a largely subjective standard with no defined measurement methodology. The federal preemption argument also matters: organisations with operations in Colorado or other states developing their own AI frameworks need to track whether state-level audit and transparency requirements survive or are displaced. More broadly, the proposal illustrates the degree to which AI governance remains contested at the highest levels of government. Security and risk teams advising on AI adoption should ensure their vendor due diligence and contractual frameworks account for a fast-shifting regulatory environment, and that AI-use documentation is maintained should compliance inquiries emerge.
What to do now
- Monitor the FTC’s policy statement process and any formal rulemaking that follows, particularly if your organisation uses or resells AI systems marketed on reliability or objectivity claims.
- Review vendor contracts and AI tool procurement documentation to understand how suppliers characterise model accuracy and neutrality, given these representations may become material to FTC scrutiny.
- Track the status of the Colorado AI Act’s bias audit requirements, as the federal preemption question is unresolved and state obligations may still apply to organisations deploying models in that jurisdiction.
- Ensure internal AI governance documentation captures how models were selected, what disclosures were reviewed, and how bias and accuracy risks were assessed — this creates a defensible record regardless of which regulatory framework ultimately prevails.
- Engage legal and compliance advisers on the First Amendment and consumer protection dimensions of this proposal before making commitments to AI system configurations or outputs in customer-facing applications.
