AI Regulation Tracker / Enforcement and deception
The FTC Made a Facial Recognition Vendor Prove Its Bias-Free Claims or Stop Making Them
A finalized consent order, effective January 2025, that sets the US substantiation bar for AI accuracy and bias marketing. The Federal Trade Commission finalized an order against IntelliVision Technologies over claims that its facial recognition software was one of the most accurate on the market, performed with zero gender or racial bias, was trained on millions of faces, and could not be fooled by a photo or video. The order bars those claims unless the company holds competent and reliable testing to back them.
This one is a few months old on the calendar, and it matters more now than it did the day it landed. The Federal Trade Commission finalized an order against IntelliVision Technologies Corp. in January 2025, closing out a case it had first announced the previous December. IntelliVision sells facial recognition software that runs inside home security systems and smart home touch panels. The problem was not the software. The problem was what the company said about it.
The FTC's account is blunt. In the agency's words, IntelliVision "did not have evidence to support its claims that its software has one of the highest accuracy rates on the market and performs with zero gender or racial bias." That is the whole case in one sentence. The company marketed a bias-free, best-in-class product and, according to the Commission, could not produce the testing to stand behind it. Under Section 5 of the FTC Act, that is a deceptive act, and it does not matter that the product was AI.
The claims the FTC went after
There were four, and they are worth naming because they are the exact claims US vendors make every day. First, accuracy: that the software was one of the most accurate on the market. Second, bias: that it performed with zero gender or racial bias. Third, training scale: that it was trained on millions of faces. Fourth, anti-spoofing: that it could not be fooled by a photo or a video image.
The training claim is the one that should make every marketing team pause. The FTC alleged that IntelliVision "did not train its facial recognition software on millions of faces, as it claimed, and instead trained its technology on images of approximately 100,000 unique individuals, and then used technology to create variants of those same images." A hundred thousand real people, augmented into a bigger-sounding number. That gap between the pitch and the record is exactly what an enforcer looks for.
What the order actually forbids
Read the operative language precisely, because the point of this order is not a fine. It is a rule about future speech. The final order bars IntelliVision from misrepresenting the accuracy and efficacy of its facial recognition software, its comparative performance across different genders, ethnicities, and skin tones, and its ability to detect spoofing. Then it goes further. The order prohibits the company from making any representation about the effectiveness, accuracy, or lack of bias of its technology "unless it possesses and relies on competent and reliable testing of the technology."
That is a substantiation-first regime. The company is not banned from ever saying its product is accurate or unbiased. It is required to hold the evidence before it speaks. The phrase competent and reliable testing is the whole ballgame. It means the proof has to exist, has to be sound, and has to be in hand at the time of the claim, not assembled later if someone asks.
Why a 2025 order is a 2026 problem for US vendors
The reason this belongs on your desk now is that it is not an isolated shot. The FTC framed IntelliVision as its second major AI facial recognition case in a year, following the Rite Aid surveillance settlement. The through-line is a policy the agency stated plainly through Samuel Levine, then Director of the Bureau of Consumer Protection: "Companies shouldn't be touting bias-free artificial intelligence systems unless they can back those claims up. Those who develop and use AI systems are not exempt from basic deceptive advertising principles."
That sentence is the standard. It applies to any US business, not just facial recognition vendors. If your website says your AI is unbiased, fair, the most accurate, or impossible to fool, the FTC's position is that you are making an advertising claim and you carry the burden of proof. The technology being novel is not a defense. The claim being aspirational is not a defense. The order treats "unbiased AI" the same way it would treat "clinically proven" on a supplement label.
What to do with this
For US companies and their counsel, the move is a claims audit, and it is cheap relative to a consent order. Pull every public statement your company makes about the accuracy, fairness, bias, or security of any AI system. For each one, ask a single question: do we hold competent and reliable testing that supports this exact claim, and can we produce it today. If the answer is no, either commission the testing or soften the claim to what the evidence supports. Vague comparative superlatives like most accurate and words like zero bias are the highest-risk phrasing, because they invite the precise scrutiny IntelliVision could not survive.
Counsel should also look past first-party marketing. Read your vendor contracts and your resellers' materials, because claims you repeat or pass through can become your exposure. IntelliVision's software was sold into third-party home systems, and the FTC still went to the developer of the claims. If you are the source of an accuracy or bias representation that others amplify, you own it. The defensible posture is documented substantiation held before the claim goes out, refreshed as the model changes, and mapped to the specific words on the page.
Questions professionals are asking
What did IntelliVision actually get in trouble for?
Not the software itself, but the marketing. The FTC alleged IntelliVision had no evidence for claims that its facial recognition software was one of the most accurate on the market, performed with zero gender or racial bias, was trained on millions of faces, and could not be fooled by a photo or video. Under Section 5 of the FTC Act, unsupported claims like these are deceptive.
Is this a law or a one-off settlement?
It is a finalized consent order in a single case, but it carries the force of law against IntelliVision for 20 years and each violation can draw a civil penalty of up to 51,744 dollars. As precedent it signals how the FTC applies long-standing deceptive-advertising principles to AI marketing generally.
What does competent and reliable testing mean?
It is the substantiation standard the order imposes. Before IntelliVision makes any claim about the effectiveness, accuracy, or lack of bias of its technology, it must possess and rely on competent and reliable testing that supports that claim. The evidence has to exist and be sound at the time the claim is made, not gathered afterward.
Does this apply to companies that are not facial recognition vendors?
The order binds IntelliVision, but the principle reaches any US business. The FTC's stated position is that companies should not tout bias-free AI unless they can back it up, and that AI developers and users are not exempt from basic deceptive-advertising rules. Any accuracy, fairness, or bias claim about an AI product is an advertising claim you must be able to substantiate.
What should our legal and marketing teams do now?
Run a claims audit. Identify every public statement about the accuracy, fairness, bias, or spoof-resistance of your AI, and confirm you hold competent and reliable testing for each specific claim that you can produce on demand. Where you cannot, commission the testing or narrow the claim. Treat superlatives like most accurate and absolutes like zero bias as the highest-risk phrasing.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how the FTC Act and this order apply to your marketing with qualified counsel.