From Sidhant Tamrkar | Product & Market Analysis
FTC AI Settlements in 2026: The $930,000 Cox Case and the Claims Now Banned
On this page
On 27 August 2026 the Federal Trade Commission finalised orders requiring Cox Media Group and two partner firms to pay $930,000 for selling an "Active Listening" AI ad product that did not listen to anything. Read beside the Growth Cave and Air AI orders, it shows exactly which AI marketing claims the FTC now treats as deceptive. The list is short enough to check against your own copy this week.
Key takeaways
- The FTC resolved 3 AI-related marketing cases in 2026, all by 2-0 votes. Growth Cave in January, Air AI in March, and Cox Media Group with MindSift and 1010 Digital Works, proposed in May and finalised in August.
- The Cox order bans misstatements in 3 named categories. The qualities or features of the ad service, the collection and consent status of voice data, and geographic targeting capability. Each one maps onto ordinary B2B AI product copy.
- Partners paid too. MindSift and 1010 Digital Works, the two smaller firms that sold the product with Cox, each paid $25,000. Repeating a vendor's AI claim in your own sales material is not a safe harbour.
- Collected cash is a tiny share of the judgments. Air AI's operators pay $50,000 against an $18 million judgment, about 0.28%. Read the conduct bans, not the dollar figures, to understand the real cost of an order.
The short answer
In 2026 the FTC settled AI marketing cases against Cox Media Group, MindSift and 1010 Digital Works ($930,000 total), Air AI ($18 million judgment, $50,000 paid) and Growth Cave. The orders ban misrepresenting what an AI product does, what data it uses, whether consumers consented, and what earnings or efficiency it delivers.
What the FTC's 2026 AI settlements actually cover
Three FTC matters settled in 2026 carry an explicit AI element. Only one is a pure AI marketing case. The other two are business-opportunity schemes in which "AI" was part of the pitch, and the orders added AI-specific language to the usual bans.
FTC AI enforcement in 2026 has not targeted AI products as a category. It has targeted companies that misdescribed their AI, or promised money that AI would make for the buyer.
The FTC's own AI topic page lists the year's enforcement items in date order. Between January and August 2026 it shows the Air AI settlement in March, the proposed Cox Media Group orders in May, the July policy statement on AI accuracy, and the final Cox orders in August. Growth Cave sits on the general press release list, with a single AI sentence in its order summary.
My read is that this is the shape of FTC AI enforcement under the current Commission. Cases are few, and all are grounded in ordinary deception law. Nothing in the 2026 record suggests a new theory.
Every 2026 FTC AI order, case by case
The table below compiles every 2026 FTC order with an AI element traceable to an ftc.gov press release. The prohibited language is quoted from the FTC's own summaries, not from commentary.
| Case and date | Money | What the AI claim was | Prohibited claim language (quoted) |
|---|---|---|---|
| Growth Cave, 27 Jan 2026. Federal court, C.D. California. | Judgments of $48,597,538, partially suspended. Asset liquidation, plus $43,000 from relief defendant Friendly Solar. | "AI software" called GrowthBox that would "automate nearly 100% of the process", per DLA Piper's reading of the complaint. | Orders "prohibit the defendants from making misleading representations related to earnings claims, testimonials, and the use of artificial intelligence." |
| Air AI, 24 Mar 2026. Proposed order, D. Arizona. | $18 million judgment, largely suspended. $50,000 to the Commission. | Misrepresented "the performance, efficacy, nature, or central characteristics" of its services, and earnings potential. | Banned from "Making false claims or misrepresentations while selling any goods and services" and "Making earnings claims without adequate substantiation or disclosure". |
| Cox Media Group, MindSift, 1010 Digital Works, proposed 21 May, final 27 Aug 2026. Administrative. | $880,000 (CMG), $25,000 each (MindSift, 1010 Digital Works). $930,000 total. | "Active Listening" service that listened to conversations on smart devices to target ads. It used resold email lists. | Prohibited from "any misrepresentation about the qualities or features of its advertising or marketing services; the collection and use of voice data and whether consumers have provided their consent" and "the geographic targeting capabilities". |
Quoted language is from FTC press releases unless stated. The Growth Cave product description comes from DLA Piper's February 2026 analysis of the complaint, a Tier 2 source. Retrieved 27 October 2026.
Cox Media Group: the $930,000 Active Listening order
This is the case B2B marketers should study first, because it is a business-to-business sale. The customers were advertisers buying a targeting product, not consumers buying a course. That makes it the closest 2026 precedent for a SaaS company describing its own AI to other businesses.
The pitch and what sat behind it
The FTC's May 2026 announcement set out the pitch. The three firms "claimed their 'Active Listening' branded marketing service listened in on consumers' conversations overheard by smart devices, in real time, to target advertising". The FTC said the service "did not, in fact, listen in on consumers' conversations or use voice data at all".
What customers actually received, the FTC said, "consisted of reselling, at a significant markup, email lists obtained from other data brokers". The price premium came from the AI story, and the AI story was false.
Christopher Mufarrige, Director of the FTC's Bureau of Consumer Protection, used no AI vocabulary at all. He said: "It is a basic rule of business that you need to be honest with your customers, and these companies failed to do that". That sentence is the whole 2026 enforcement philosophy in one line.
The consent problem hiding inside the order
The firms claimed consumers had "opted in" by agreeing to terms of service. The FTC answered that "Clicking through mandatory terms of service does not constitute 'opt-in consent'."
The final order goes further. The FTC said that "if the service had functioned as advertised, this collection and use of consumers' voice data without adequate consent would itself violate the FTC Act". So Cox was exposed both ways: the claim was false, and the true version would also have been unlawful.
If your AI product rests on data collected from someone else's users, your consent claims are representations the FTC can test. The data practice and the copy are one exposure, not two. The FTC's separate interest in how vendors disclose data-driven pricing runs on the same logic, covered in the analysis of FTC personalized pricing and vendor disclosures.
The final order was approved by a 2-0 Commission vote after a 30-day comment period that drew 2 comments, per trade coverage of the final release. MindSift and 1010 Digital Works each paid $25,000, and the FTC directed the money to redress for CMG's customers.
Growth Cave and Air AI: AI claims inside business-opportunity cases
The other two 2026 FTC AI orders came out of business-opportunity enforcement. In both, AI was the hook used to sell an income promise.
Growth Cave's GrowthBox order
On 27 January 2026 the FTC announced settlements with the remaining Growth Cave defendants, including co-CEOs Lucas Lee-Tyson and Osmany Batte. The FTC alleged the scheme took "nearly $50 million" from consumers. Judgments of $48,597,538 were partially suspended, and the co-CEOs must liquidate assets including a house, a Rolls-Royce and a Ferrari.
The AI element is one clause. The FTC said the orders "prohibit the defendants from making misleading representations related to earnings claims, testimonials, and the use of artificial intelligence". DLA Piper's reading of the order text, published 5 February 2026, is that it bars misrepresenting that a product "will use artificial intelligence (AI) to maximize revenues" or enhance "profitability, effectiveness, or efficiency".
That wording matters because it is not unique to Growth Cave. DLA Piper reports the FTC used nearly identical language in orders filed in May and June 2025 against Ascend Ecom, Empire Holdings Group and FBA Machine. When the same clause appears in four orders, it is a template, and templates tell you what the agency will ask for next time.
Air AI's $18 million judgment
Air AI sold conversational AI tools and licences to entrepreneurs and small businesses. The FTC sued in August 2025 and announced a proposed settlement on 24 March 2026. The order bans Air AI and its owners, Caleb Maddix, Ryan O'Donnell and Thomas Lancer, from selling or marketing any business opportunity.
The complaint alleged the company misrepresented "the performance, efficacy, nature, or central characteristics of their services". It also alleged false claims of a "refund or buy-back guarantee" and false earnings claims. Only one is about AI capability, phrased in the most general terms the FTC has.
The ban that follows is equally broad: no "false claims or misrepresentations while selling any goods and services". For software companies, I think this matters more than any AI-specific language. A capability claim about an AI product is judged like any other product claim, with no special defence.
What the dollar figures hide about FTC AI enforcement
Headline amounts in FTC AI settlements are a poor guide to the cost of an order. Two of the three 2026 cases carried large judgments the FTC suspended on inability to pay.
Face-value judgments come to about $67.5 million: $48,597,538 for Growth Cave, $18 million for Air AI and $930,000 for the Cox group. Fixed cash payments named in the press releases come to $1,023,000: $930,000 from the Cox group, $50,000 from Air AI and $43,000 from Growth Cave's relief defendant. That is about 1.5% of face value, before counting Growth Cave's asset sales, whose value the FTC did not state.
The case with the smallest judgment, Cox, produced 86% of the fixed cash. Cox's $880,000 alone is 94.6% of the $930,000 in its own case. A solvent company with customers to keep pays in full.
For a solvent SaaS vendor, the payment is the cheap part. The expensive parts are the conduct ban, which in FTC administrative orders typically runs for years, the compliance reporting, and the customer conversations after a press release names your product.
What the 2026 FTC AI enforcement pattern says about Section 5
Section 5 of the FTC Act prohibits unfair or deceptive acts or practices. Every 2026 AI order rests on the deception prong. None relies on the unfairness theory that some 2024 cases used against AI tools themselves.
The Rytr order set aside
On 22 December 2025 the Commission reopened and set aside its 2024 consent order against Rytr, an AI writing tool that could generate reviews. Venable's analysis, published 5 January 2026, reports the vote as 2-0 and the FTC's conclusion that the original complaint "failed to adequately allege a violation of Section 5".
Chairman Andrew Ferguson had dissented from the original case. Holland & Knight quotes his dissent: "Treating as categorically illegal a generative AI tool merely because of the possibility that someone might use it for fraud" was, in his view, the error. The FTC also found the order "imposed an unjustified burden on innovation in a young and rapidly evolving AI market," per Venable.
The Rytr reversal is being over-read. It narrows liability for building a tool that someone else might misuse. It does nothing for a company that misdescribes its own tool, which is exactly what the 2026 orders punish.
The July 2026 AI accuracy policy statement
On 1 July 2026 the FTC voted 2-0 to seek comment on a proposed policy statement on AI accuracy. It addresses concern that AI companies "may be manipulating the behavior of their AI systems contrary to reasonable consumer expectations for objectivity and accuracy". Comments closed on 31 July 2026.
The statement is a proposal and creates no new obligation. Its interaction with state law is covered in the analysis of the FTC's Section 5 theory against Colorado's AI law. The FTC is signalling that undisclosed changes to how an AI system behaves can be deceptive, which makes accuracy claims in your copy a live surface.
Holland & Knight's August 2026 summary captures the net effect: "reduced enforcement directed at AI innovation itself, coupled with continued enforcement against deception". The safe zone for building AI widened in 2026. The safe zone for describing it did not move.
Seven rules for B2B AI marketing compliance, drawn from the orders
The orders translate into rules for anyone writing AI product copy, sales decks or partner enablement. Each rule below traces to specific order language. The example phrasings are illustrative, written by us, and are not quoted from any company.
| Rule | Weak version (illustrative) | Version that holds up | Order it comes from |
|---|---|---|---|
| 1. Describe the mechanism, not the brand name | "Our AI listens for buying signals" | "We match accounts against third-party intent data from named providers" | Cox: qualities or features of the service |
| 2. State the data source and consent basis | "Users opted in via our terms" | "Data comes from partners who collect express opt-in consent; ask for the list" | Cox: voice data and consent |
| 3. Prove any targeting or precision claim | "Reach buyers in any neighbourhood" | "Targeting resolves to postcode level in our tests; methodology on request" | Cox: geographic targeting capability |
| 4. Do not claim an automation share you have not measured | "Automates nearly 100% of the process" | "Automates drafting and routing; your team still uploads assets and approves sends" | Growth Cave: use of artificial intelligence |
| 5. Substantiate every revenue or efficiency outcome | "AI that grows revenue 3x" | "Median customer cut handling time by a stated amount, sample and period disclosed" | Air AI and Growth Cave: earnings claims |
| 6. Treat partner and reseller copy as your copy | Letting resellers improvise capability claims | Approved claim library that partners must use verbatim | Cox: MindSift and 1010 Digital Works paid |
| 7. Keep testimonials and guarantees literal | "Guaranteed results or your money back" | Refund terms written once, honoured exactly as stated | Air AI: refund or buy-back guarantee |
Rule 5 is where most B2B AI copy fails today. A capability claim with no stated sample is the same object as an earnings claim without "adequate substantiation", the Air AI order's phrase. Customer outcome numbers deserve the same treatment, and the method for checking them is set out in the guide to case studies with verified outcomes.
Rule 6 is the one most likely to surprise a sales leader. The two smaller firms in the Cox case were not the product owner, yet each paid $25,000 and each is bound by the same conduct ban. If a reseller pitches your AI with claims you did not approve, the exposure travels in both directions.
The SEC runs a parallel campaign against public companies that overstate AI, logged in the SEC AI-washing enforcement log. The underlying test is identical: say what the system does, and be able to show it.
The rules also cut against the instinct to stay vague to stay safe. "AI-powered" with nothing behind it is the Cox fact pattern in miniature. Specific claims with evidence are both more persuasive and easier to defend, a point the data in the analysis of AI disclosure and customer trust supports from the buyer side.
Where this reading of FTC AI enforcement is weakest
This post builds a rulebook from three cases, which is a thin base.
Three cases is a small sample, and only one is B2B
Two of the three 2026 orders arose from business-opportunity schemes aimed at individuals hoping to earn income. Applying their AI clauses to enterprise SaaS copy is an inference, not a holding. A case against a mainstream B2B software vendor could draw lines differently.
The Cox case is the strongest precedent because it is business to business. It is also an extreme fact pattern: the FTC alleged the core feature did not exist. Most vendors overstate a real capability rather than inventing one, and no 2026 order tests where overstatement becomes deception.
Press release summaries are not order text
The prohibited language quoted in this post comes from FTC press releases and, for Growth Cave, from DLA Piper's reading of the order. We could not extract the full text of the Growth Cave and Air AI order PDFs during research. Order definitions and carve-outs can narrow or widen a ban in ways a summary omits.
Workado, an AI-detection vendor, has a final FTC order that requires it to "refrain from making any representations that are false, unsupported or misleading," per Holland & Knight. Sources disagree on when it became final, so it is excluded from the 2026 table rather than placed on a guessed date.
Finally, the FTC is one regulator. State attorneys general, the SEC and private plaintiffs all police AI claims under their own laws. A clean read of the FTC record is not a clean bill of health.
Frequently asked questions
What AI settlements did the FTC announce in 2026?
Three AI-related matters settled in 2026. Growth Cave in January, with judgments of $48,597,538 partially suspended. Air AI in March, with an $18 million judgment largely suspended and $50,000 paid. Cox Media Group, MindSift and 1010 Digital Works, proposed in May and finalised on 27 August 2026, paying $930,000 in total over a fake "Active Listening" AI ad product.
What did Cox Media Group's Active Listening product claim to do?
The firms claimed the service listened to conversations picked up by smart devices, in real time, to target ads to consumers in a specific area. The FTC said it did not use voice data at all. It resold email lists from other data brokers at a significant markup. The firms also claimed consumers had opted in through terms of service, which the FTC rejected.
What AI marketing claims does the FTC prohibit?
The 2026 orders prohibit misrepresenting the qualities or features of an AI product, how it collects and uses data, whether consumers consented, and its targeting capability. They also ban misleading claims that a product uses AI to raise revenue, profitability, effectiveness or efficiency, and earnings claims without adequate substantiation. The common test is whether the claim is true and provable.
Does the FTC still enforce against deceptive AI claims after Rytr?
Yes. The FTC set aside the Rytr order in December 2025 because it found the complaint did not allege a Section 5 violation and the order burdened innovation. That reversal concerned liability for a tool someone might misuse. The 2026 orders show the FTC still pursues companies that misdescribe what their own AI products do.
Can a reseller be liable for a vendor's false AI claims?
The Cox case suggests yes. MindSift and 1010 Digital Works sold the Active Listening service alongside Cox Media Group. Each paid $25,000 and each is bound by the same prohibition on misrepresenting the service's features, voice data use, consent and geographic targeting. Partners who repeat capability claims should verify them before using them.
How should a B2B company check its AI marketing for FTC compliance?
List every claim about what your AI does, what data it uses, and what results it delivers. For each, record the evidence, its date, and who owns it. Remove any claim without evidence, replace brand language with a description of the mechanism, and give partners an approved claim library. Recheck the list whenever the product changes.
How to audit your AI copy against the orders this week
Pull your homepage, your main AI product page and your most-used sales deck. Highlight every sentence that says what the AI does, what data it uses, or what result it produces. In most B2B companies that is fewer than 30 sentences, and it takes an afternoon.
Then put each sentence through one question from the Cox order: could you show a customer, today, the evidence that it is true? Anything that fails gets rewritten into a mechanism description or cut. Send the surviving list to your partners as the only approved wording, and date it, because the next product release will make some of it false.
Use the seven rules
Run your AI product page against the rules table above before your next launch. If you sell through comparison pages, pair it with the B2B comparison page template.
References
- Federal Trade Commission, FTC Finalizes Orders with Cox Media Group, Two Other Firms Settling Charges They Deceived Customers About "Active Listening" AI-Powered Marketing Service, 27 August 2026. Used for final amounts, vote and prohibited claim categories.
- Federal Trade Commission, FTC to Require Cox Media Group, Two Other Firms to Pay Nearly $1 Million, 21 May 2026. Used for the product description, opt-in language and Mufarrige quote.
- Federal Trade Commission, Air AI and its Owners will be Banned from Marketing Business Opportunities, 24 March 2026. Used for the judgment, payment, allegations and bans.
- Federal Trade Commission, FTC Secures Settlement Banning Growth Cave Defendants, 27 January 2026. Used for judgments, asset terms and the AI clause.
- Federal Trade Commission, FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy, 1 July 2026. Used for the proposed policy statement and comment deadline.
- DLA Piper (Tobey, Carr, Atleson), FTC resolves another case involving AI-washing, 5 February 2026. Used for Growth Cave order wording and the 2025 template orders.
- Holland & Knight (DiResta), A Look at the FTC's Current Approach to AI Regulation, 19 August 2026. Used for the Rytr date, Ferguson dissent, Workado and the policy summary.
- Venable, All About Advertising Law, The FTC Walks Back Its Rytr Enforcement Action, 5 January 2026. Used for the Rytr vote and reasoning.
Weakest point in the source base: order wording is quoted from FTC press releases and law-firm summaries, not from the full order PDFs, which could not be text-extracted during research. Last verified 27 October 2026.
Related reading