FTC Floats AI Policy Aiming To Ensure That AI Makers Disclose The Truth About Biases In Their LLMs

Trend Of Laws Seeking To Regulate AI

Before we explore the FTC proposal, I’d like to set the stage by laying out the national AI governance landscape in a nutshell. The efforts of the FTC are inextricably intertwined with the larger mosaic of new AI laws and legal undertakings at the federal and state levels across the United States and internationally, too.

A beehive of activity is underway to craft new AI laws. Individual states have been pushing ahead with establishing new AI laws. Some of these AI laws are similar, while others differ and even conflict with those of other states. See my comprehensive coverage of the many newly enacted AI laws at the link here.

Using laws to control AI is a matter on the minds of the public and in the hands of legislators. Some ardently believe that AI and AI makers are being allowed to run amok. New AI laws are vitally needed to protect society from this onslaught of ubiquitous AI. Not everyone agrees with this pell-mell rush of new AI laws, or at least they are concerned that these AI laws might go overboard. In the zeal to protect society, there is a chance that we might unduly restrict innovation and delay or undercut the benefits of leading-edge AI.

The debate is ongoing and heated.

Readers might recall that I proposed a 7-step AI-law-making process that I believe could substantively help regulators to devise new AI laws that are on target and balanced; see my depiction at the link here. This has the added benefit of reducing what I refer to as AI-law legal debt. This refers to AI laws that, though they look pristine, contain hidden legal debt that must ultimately be paid. Legal debt consists of glitches and hitches that will eventually be found when AI laws have been enacted without sufficient scrutiny and analysis (see my description of AI legal loopholes at the link here, and a list of the common blunders in AI laws at the link here).

The Budding Morass Of AI Laws

In terms of the AI laws in the United States, none of these new AI laws have yet fully stood the test of time, meaning that we won’t really know how well they stand up until there are court challenges that test these new laws. It is too early to know whether these laws being rapidly passed will survive legal battles waged by AI makers and other contenders. Just because AI laws are enacted does not mean they are legally proper. All sorts of improper provisions and constitutionally contentious stipulations are undoubtedly buried within these sparkly new AI laws.

Congress has repeatedly waded into establishing an overarching federal law that would encompass AI. So far, no dice. The efforts have ultimately been stymied. At this time, there isn’t an overarching federal law devoted to a comprehensive set of AI considerations. Some ardently believe that the need for a big-bang federal law on AI is obvious, while others express concern that any such federal law would be onerous and a fitful one-size-fits-all construct.

The crux is that there is intense and pervasive interest in using the law to govern AI, but getting to a juncture of devising AI laws that are legislatively passable and real-world implementable is a notable sticking point. It is an abundantly mushrooming realm. AI makers would be wise to keep a close eye on what is happening in the hallways and byways of regulators and legislative bodies. I have repeatedly noted that a valuable specialty for budding lawyers is to consider concentrating on the exciting and dynamic field of AI and the law; see my predictions and recommendations at the link here.

Worrying About The Problems Of AI

When it comes to generative AI and LLMs, there are four major concerns about the pitfalls of today’s capabilities:

• (1) Errors by AI.
• (2) Falsehoods produced by AI.
• (3) AI hallucinations of AI.
• (4) Biases within AI.

For my in-depth review of those four concerns, see the link here. Let’s take a brief look at the AI bias concerns since that’s most pertinent to the FTC proposal that I will be unpacking shortly.

Biases within AI can arise in a multitude of ways. Do not fall for imprudent claims that there is somehow only one means by which AI can end up containing biases. There are plenty of avenues.

Consider these persistent possibilities:

• Biases in the scanned data from the Internet that was used for data-training of the AI.
• Biases in the AI algorithms used to pattern-match on the sourced data.
• Biases in the overall AI design and its infrastructure.
• Biases of the AI developers, either implicitly or explicitly, in the shaping of the AI.
• Biases of the AI testers, either implicitly or explicitly, in the testing and refinement of the AI.
• Biases of the RLHF (reinforcement learning by human feedback) tuning efforts, either implicitly or explicitly, via the human reviewers imparting training guidance to the AI.
• Biases of the AI fielding facilitation for the operational use of the AI.
• Biases in any setup or default system prompts or system instructions established for the AI in its daily usage.
• Biases introduced or expanded during AI maintenance or upkeep of the AI.
• Etc.

For anyone expecting that a magic wand is going to cure the plethora of potential biases, please set aside that dreamy notion. It is going to be a hard and enduring battle.

FTC Looking At AI Deceptive Conduct

Shifting gears, we are now ready to dive into the FTC’s new proposed initiative.

An overarching mission of the Federal Trade Commission (FTC) is to prohibit businesses from engaging in unfair or deceptive conduct. This oversight is typically undertaken in crucial ways, such as encouraging active competition in the marketplace, educating consumers, detecting and dealing with false advertising, and so on. Among the many safeties, the FTC tries to surface deceptive conduct and accordingly rein it in.

In the realm of AI, the FTC is now aiming to tackle the issue of AI that produces less-than-truthful outputs. The FTC asserts that this is likely a deceptive practice of notable magnitude and therefore falls within the contours of the FTC. The FTC indicates that the public has come to rely upon AI on a massive scale and that the assumption by consumers is that AI is producing truthful outputs. This public assumption is readily spurred by how AI makers tout their AI wares.

If the AI that consumers are using contains biases that lead to less-than-truthful outputs, consumers are being deceived. How do these deceptions arise? It could be by AI makers without any intention on their part, though it nonetheless could be occurring anyway. It might also be undertaken by AI makers who have either a hidden or an outright intention. Some AI makers could be trying to insidiously achieve greater usage, consumer loyalty, and monetization, and bake biases into the AI to meet those objectives. Some AI makers could be doing so without conscious awareness of their actions, but their AI developers are inherently allowing their personal biases to permeate the design and shaping of the AI.

A brand-new additional twist might catch you by surprise.

The AI might be seeded with biases due to compliance with emerging state-level AI laws. Here’s the deal. An AI maker might believe they must bias their AI to comply with some particular state-level legal stipulation. If they don’t do so, they risk the wrath of state-level authorities coming after them legally. You might say that an AI maker could end up between a rock and a hard place. It goes this way. They could opt to not allow their AI to be used in such a state, but that cuts down on their base of potential users and creates a tough choice of what to do.

FTC Makes A Proposal

The FTC posted a proposal entitled “Federal Trade Commission’s Proposed Policy Statement Concerning The Suppression Of Accuracy In Artificial Intelligence Systems” on July 1, 2026, which contains these salient points (excerpts):

• “AI companies have spent years representing explicitly and implicitly that their systems aim to produce the best output — output that faithfully and accurately achieves users’ stated objectives and the built-in objectives that users expect in the AI system — that is possible within their technological and resource constraints.”
• “Because of these representations and the inherent nature of the products and services in question, consumers have a reasonable expectation that AI systems aim to give truthful and accurate outputs.”
• “Nonetheless, an AI company might be tempted to alter or steer the output of its systems contrary to consumers’ reasonable expectations for various reasons.”
• “Section 5 of the FTC Act’s prohibition on deceptive acts or practices secures consumers’ right to truth in the marketplace.”
• “In light of the above, the Commission believes that AI companies that steer the outputs of their AI systems toward unexpected objectives, and away from the objectives set by or reasonably expected by users, are likely to deceive consumers in violation of Section 5 of the FTC Act.”

You can see that the FTC is asserting that consumers are potentially being duped by AI that is generating less-than-truthful outputs. The FTC contends that this triggers its involvement based on Section 5 of the FTC Act.

As I mentioned earlier, the public can respond to the proposal by providing written feedback by July 31, 2026. Anyone steeped in AI, including and perhaps especially AI ethicists and AI legal experts, should consider undertaking a careful reading of the document and providing a response if they have something demonstrative to say about the proposal.

Analysis Of The Approach

A supporting perspective is that the FTC has touched on something that has been brewing and will finally receive the attention it deserves. People expect AI to be truthful. They don’t expect AI to answer questions unevenly. If an AI is going to proceed down the uneven route, the AI ought to at least clearly alert the user.

In that sense, the FTC is not saying that the AI cannot be biased, but that if it is biased, the AI must provide sufficient and due notice to the consumer. This mirrors traditional FTC doctrine. If consumers reasonably expect a product to possess a characteristic, a company generally cannot quietly alter that characteristic while continuing to market the product as though nothing has changed.

Under this approach, the central question is not whether an AI system is objectively biased, but whether consumers are being misled about how the AI behaves.

Clarifying The Scope

I’ve observed that some have seemingly misinterpreted the proposal. They appear to interpret the FTC as saying that AI cannot be biased. The worry is that the FTC is going to force AI makers into a claimed “unbiased” decree. I don’t believe that’s what the case is.

My reading is that if the AI does not genuinely attempt to provide truthful answers, the AI maker cannot allow users to believe that it does, as stated this way in the proposal: “Of course, a company may be able to avert potential deception by making truthful, non-misleading representations about the aims of its model. But such representations would need to make clear that the AI company is prioritizing objectives different than those consumers requested or would otherwise expect.”

I realize that some might believe this position is a wink-wink and a slippery slope. They envision that the FTC might harangue AI makers into declaring disclosures of all manner of AI outputs. This, in turn, would sneakily steer the AI makers toward ensuring that the AI outputs do not become disclosure-worthy. To what end might that pinch AI?

This is likely the kind of feedback that will be provided regarding the proposal.

Analogy And Proportionality

Suppose an AI maker modifies its LLM to favor one selected viewpoint, systematically omitting certain facts, exaggerating selected evidence, or steering users toward predetermined conclusions. The FTC’s concern would likely be less about the viewpoint itself and more about whether consumers understand that this is occurring. The same reasoning applies if an AI has systematic commercial biases. For example, preferentially recommending products from paying partners while presenting the recommendations as a purely objective analysis.

The deception arises from the mismatch between actual operation and consumer expectations. The proposal also encompasses the role of proportionality. The more substantial the behavioral departure, the more prominent the disclosure would likely need to be.

The State-Level Consideration

I have also heard some claiming that this is going to create a head-to-head conflict between what some of the states stipulate in their new AI laws and what the FTC is telling AI makers they must do on a federal basis. This seems to create two separate and loggerhead legal duties. How is an AI maker to choose between the two?

Again, I don’t see it that way. The usual situation is that state AI law governs how the AI must behave. The FTC proposal governs what consumers must be told about that behavior. These obligations are not necessarily inconsistent. An AI maker can comply with both by implementing the legally required behavior while accurately informing users that the behavior exists.

The exception would be if a state AI law ventured into what the AI is to tell the user about the behavior, such as legally restricting the AI maker from being upfront about the matter or disallowing disclosures. This would likely give rise to a legal contention of a nature that would be placed before the courts.

Devil In The Details

Those who have raised their eyebrows about the FTC proposal will seemingly offer a variety of qualms. What does it mean to say that an AI is less-than-truthful or inaccurate? Who determines this? Without a clear-cut set of boundaries and metrics, an AI maker would presumably be in the dark about whether they need to alert consumers about the behavior or whether it is okay to not mention it. They could get on the wrong side of the matter and not realize they have tread into hot water.

The same difficulties come into play when the manner of disclosure is considered. Should the AI ring bells and flash bright lights at the user? What is the expected extent of the disclosures that need to be made? Can the disclosures be made so that users don’t get exasperated and frustrated by the AI notifications? The deployment and enforcement elements of the policy are going to be complicated.

The FTC will need to establish workable standards for determining the “reasonable consumer” baseline, identifying which behavioral differences are sufficiently material to require disclosure, specifying what constitutes a clear and conspicuous disclosure for generative AI, and addressing how disclosures should evolve as AI systems are continuously updated. That’s a mouthful.

The World We Are In

AI is becoming ubiquitous. Governing AI is not going to be easy. People have dramatically different opinions on dealing with AI and AI makers. Trying to arrive at appropriate compliance measures, safety guardrails, civil-rights protections, and content moderation is going to test our societal mettle.

No ready-made answers are sitting around. Lots of challenging work is ahead. I hope that we are willing to listen to viewpoints across the board. As Henry Ford famously said: “Coming together is a beginning; keeping together is progress; working together is success.”