A. What Biden Started and Trump Stopped

(1) The Biden Administration's Push to Regulate AI

On the morning of September 25, 2024, FTC Chair Lina Khan appeared at the press room of the Federal Trade Commission headquarters in Washington, D.C. She stood before the microphone and declared war in a single sentence: "It is illegal to use AI tools to deceive, mislead, and defraud people."

Khan's words were plain. But behind them lay months of investigation and thousands of pages of evidence. The operation she announced was called "Operation AI Comply." It sounded like a military campaign to dismantle a drug cartel, but the targets weren't armed gangs. The targets were companies that slapped the three letters "A-I" onto their products and misled consumers.

Why was this operation necessary? The answer lay in the flow of capital.

Starting in 2023, a significant share of venture capital began pouring into AI-related companies. Wall Street investors opened their wallets the moment a company had "AI" in its name. The industry called this phenomenon "AI Washing," the digital cousin of "greenwashing," where companies with no real environmental credentials pretend to be eco-friendly.

Whether the company had actual working AI technology didn't matter. A convincing package was enough.

To understand the FTC's enforcement philosophy, you need to know what the legal term "deception" means. Think of it like a grocery label. If the package says "organic" but the produce inside was conventionally grown, the consumer paid money based on false information. The moment trust breaks, harm occurs. AI works the same way. If you bought a product because it said "powered by cutting-edge artificial intelligence," but it turned out to be nothing more than a basic script, that is deception. The FTC under the Biden administration established three principles for addressing this kind of deception.

First, stronger substantiation requirements. If a company claims its product is "AI-powered" or "algorithm-optimized," it must have scientific, objective evidence to back that claim before making it.

Second, surveillance of automated bias. If an AI produces outcomes that disadvantage certain races or genders in critical decisions like lending, housing, or employment, that constitutes an unfair practice.

Third, clear accountability. The excuse that "the black-box nature of the algorithm made the outcome unpredictable" would not be accepted.

Khan's approach resembled preemptive prevention. Her position was that the mere possibility of causing substantial harm to consumers could trigger regulatory action. This was an expansive reading of Section 5 of the FTC Act, the existing consumer protection statute, adapted for the AI era. Even if a technology had not yet been misused, the act of providing a tool capable of misuse was itself subject to challenge.

In the first announcement of Operation AI Comply, the FTC simultaneously revealed enforcement actions against five companies. Among them were a company marketing itself as a robot lawyer, a company selling fake review generation tools, and companies that used AI to promise passive income while defrauding investors. This was not a simple crackdown. It was a warning shot fired at the entire market.

The Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (EO 14110), signed by President Biden in October 2023, provided the foundation for this regulatory stance. The order gave federal agencies specific directives for AI governance and granted the FTC active enforcement authority to ensure fair competition and consumer protection in the AI market. Chair Khan used that authority to its fullest extent.

The FTC's offensive was coordinated with other agencies, including the Department of Justice (DOJ) and the Equal Employment Opportunity Commission (EEOC). It was part of a government-wide regulatory drive. But the approach was a double-edged sword. Alongside praise for imposing order on the early AI market and sounding an alarm against techno-utopianism, critics argued it was stifling the growth of innovative startups. In parts of Silicon Valley, frustration boiled over: "Washington is killing innovation."

This raises a critical question. Was the regulator trying to catch fraudsters, or was it trying to rein in the speed of technological progress itself? The answer to that question would emerge in a completely different form under the next administration.

(2) The Trump Administration's Policy Reversal and Its Impact

In January 2025, the Trump administration took office and the AI regulatory landscape flipped 180 degrees. The new administration's message was clear.

The Biden-era "preemptive regulation" was an unnecessary barrier threatening America's AI technology dominance. Immediately after inauguration, President Trump enacted a series of measures revoking or substantially revising Biden's AI-related executive orders.

The effect of the change in power was immediate inside the FTC. Lina Khan stepped down as chair, and the conservative-leaning Andrew Ferguson was installed as the new chair. Ferguson had served as an FTC commissioner under the previous administration, where he had repeatedly voted against Khan's regulatory policies. His philosophy was straightforward: "Punish fraud committed using technology, not the technology itself."

In July 2025, the White House released the "America's AI Action Plan." Its core directive was to review and withdraw federal regulations and procurement barriers deemed to impede AI innovation. The document contained a direct reference to the FTC: an instruction to review all investigations launched during the Biden era and "ensure that liability theories that unduly burden AI innovation are not pursued."

What did this mean in practice?

The regulatory blade had changed direction. In the past, broad investigations into AI models' societal impact or the fairness of data training were on the table. After 2025, the focus narrowed to specific, provable "false claims." The central question shifted from "Did the AI discriminate?" to "Does the AI work as advertised?"

But there is an important point. Operation AI Comply itself was not abolished. The Trump administration's FTC continued aggressive enforcement against clear-cut "fraud," cases where consumers were directly deceived and suffered financial harm. In some areas, enforcement actually intensified. Throughout 2025, the FTC kept issuing sanctions against AI-branded investment scams, fake review generation, and false performance advertising.

This fact drew attention among legal professionals. According to one law firm's analysis, "Speculation that this enforcement priority would disappear with the change of administration proved unfounded." This is supported by the fact that most actions under Operation AI Comply were approved with votes from both Democratic and Republican commissioners. Basic oversight of clear consumer deception was, it turned out, a bipartisan consensus.

The most dramatic example of the policy shift came on December 22, 2025. The FTC voted to withdraw the consent order it had entered against Rytr LLC in 2024. Chair Ferguson stated that "the original complaint failed to meet the legal requirements of the FTC Act, and the order unduly burdened AI innovation." This was the first instance of implementing Trump's AI executive order and the America's AI Action Plan.

On September 25, 2024, a vote took place at the Federal Trade Commission headquarters in Washington, D.C. The matter before the commission was a small AI company called Rytr LLC. The vote was 3 to 2. A narrow margin.

One of the two dissenters was Commissioner Andrew Ferguson. In his dissenting statement, he wrote: "This order risks turning honest innovators into lawbreakers and strangling a potentially revolutionary technology in its cradle."

One year later, Ferguson became the FTC Chair. And exactly what he had predicted came to pass, only in the opposite direction. It was not the technology that got strangled. It was the FTC's own order.

Let me explain what Rytr was.

Rytr is an AI writing assistant service. Users enter a few keywords and the AI writes text for them. Blog posts, emails, ad copy. Nothing unusual so far. The problem was a feature called "Testimonial & Review." Here is how it worked.

A user enters the keywords "moving company," "friendly," and "recommend." The AI then generates a review that reads as though a real customer wrote it: "I used this company for my move last year, and the staff was incredibly friendly. They moved every piece of furniture without a single scratch. I'll definitely use them again."

The problem is that the person who generated this review never actually used a moving service. It is a fabrication produced by AI. One user generated more than 83,000 moving company reviews using this feature.

The FTC under Chair Lina Khan brought two charges.

First, "Unfair Practice." In plain terms, polluting the marketplace. When fake reviews flood in, consumers can no longer distinguish real reviews from fake ones. Businesses that operate honestly suffer the consequences.

Second, "Providing Means and Instrumentalities of Deception." This concept needs some explanation. An analogy helps. If someone counterfeits money, that is a crime. But what about selling a counterfeiting machine? You didn't commit the crime directly, but you supplied the tool for it. The FTC argued that Rytr played exactly that role.

On December 18, 2024, the final consent order was issued. Rytr was prohibited from advertising, marketing, or selling any service that generates reviews or testimonials for the next 20 years.

The story could have ended there. A small AI company got cut by the regulatory blade. That's it. But American politics doesn't work that simply.

On January 20, 2025, Donald Trump became president again.

Three days later, on January 23, he signed an AI executive order. The core message: "Withdraw existing policies and guidance that obstruct American AI innovation." That same year in July, the White House published the "America's AI Action Plan." The plan gave the FTC a specific mandate: "Review all FTC final orders, consent decrees, and injunctions, and modify or withdraw those that unduly burden AI innovation."

It did not take Chair Ferguson long to find a candidate for review. There was the very case he had voted against one year earlier.

On December 22, 2025, the FTC withdrew the consent order against Rytr. The vote was 2 to 0. No dissents. The reason was simple: only two commissioners remained, and both were Republicans.

Christopher Mufarrige, Director of the FTC's Bureau of Consumer Protection, said in the press release: "Condemning a technology or service simply because it could potentially be misused is inconsistent with law and ordered liberty."

The withdrawal order ran six pages. Its core reasoning rested on two points.

First, the original complaint failed to meet the legal requirements. The complaint stated that Rytr users had generated tens of thousands of reviews. But a crucial fact was missing: there was no evidence those reviews had actually been posted online. A tool capable of creating fake reviews and the act of actually posting fake reviews are different matters. The former is a potential risk; the latter is actual harm. The FTC Act requires "actual harm or the likelihood of harm." The complaint contained no evidence of actual harm.

Second, the misuse of the 'means and instrumentalities' doctrine. Traditionally, this doctrine applied when a company directly made false statements and conveyed them to others. But Rytr never made any false statements itself. Even if AI-generated reviews contained falsehoods, that was the user's choice. A pencil company isn't held liable when someone uses a pencil to write a fraudulent document.

There's an irony here.

On the same day the FTC withdrew the Rytr order, it sent out ten warning letters related to fake reviews. Thirty minutes apart. Coincidence? The message was clear: We haven't given up on cracking down on fake reviews. We've just changed direction. We're going after companies that misuse the tool, not companies that make the tool. Chair Ferguson's position was this: we won't ban AI technology itself; we'll only punish those who actually use AI to deceive consumers.

The DoNotPay case, which the same chair approved in 2024, makes this distinction clearer. DoNotPay advertised its AI as a 'robot lawyer.' It wasn't one. That was plain deception. A false claim about capability. Ferguson voted in favor of enforcement in that case.

See the difference? DoNotPay lied about what its AI could do. Rytr told no lies at all. It just built a tool that someone could use to write lies.

This case raises a fundamental question about AI regulation.

Technology is neutral. A knife can be used for cooking or for crime. A printer produces legitimate documents and counterfeit bills. An AI writing tool drafts blog posts and fake reviews. Should the technology itself be banned, or should only its misuse be punished?

The Biden administration's FTC leaned toward the former. If a technology is highly likely to be misused, it should be prohibited from the start. The Trump administration's FTC leans toward the latter. If a technology has legitimate uses, allow it, and punish only those who actually break the law.

Which side is right? That question remains unanswered. One thing is certain: opposite judgments were issued on the same conduct within a single year. That tells us a great deal about the consistency of American AI regulatory policy.

There's something to add.

The Rytr order withdrawal was the first instance of implementing the Trump administration's AI Executive Order and AI Action Plan. First doesn't mean last. Other FTC orders could face review under the same logic: other cases from Operation AI Comply, and earlier AI-related enforcement actions. One legal commentator wrote: '2025 is the year with the most order withdrawals in FTC history.' The winds of change are blowing hard. Where those winds are headed, nobody yet knows.

B. The Fall of the Robot Lawyer

(1) DoNotPay: Sanctioning Robot Lawyer Hype

In 2015, a young British man opened his laptop in a Stanford University dormitory. Joshua Browder. He'd gotten so many parking tickets that while researching how to contest them, he decided to just build a chatbot. 'As a young person, I couldn't afford these tickets. So I became a legal expert on every possible reason to get out of a parking ticket.' That's what he said in an interview years later.

Browder was the son of hedge fund manager Bill Browder. In Silicon Valley, his story was packaged as the classic boy-genius narrative. Soon he was selected for Peter Thiel's startup fellowship and dropped out of college. He received investment from prominent venture capital firms like Andreessen Horowitz and Coatue Management. The company that started as a parking ticket chatbot quickly grew its ambitions. Its name was DoNotPay. The slogan was bolder still: 'The world's first robot lawyer.'

Browder's promises were grand. For a $36 monthly subscription, AI would handle everything from contesting highway speeding tickets to drafting divorce papers to filing small claims court petitions. The website quoted the Los Angeles Times: 'What this robot lawyer can do is remarkably similar to, or exceeds, what human lawyers do.' For consumers frustrated by the high cost of legal services, this sounded like a dream.

But the FTC's gaze was fixed on the gap between advertising copy and actual functionality. In September 2024, the FTC released its findings.

DoNotPay had never subjected its AI technology to any objective verification of whether it could perform at the level of a real lawyer, or whether the documents it generated were legally valid. The company had never hired attorneys internally to review the AI's output. The problem was straightforward: they called it a 'robot lawyer,' but no lawyer-level verification had ever taken place.

Here it's worth revisiting the legal meaning of 'deception.' When consumers buy a product, they trust the advertising. They pay money based on that trust. If the advertising doesn't match reality, consumers have made financial decisions based on false information. Section 5 of the FTC Act defines exactly this situation as 'unfair or deceptive acts or practices.'

According to the FTC complaint, DoNotPay's technology was not trained on a complete, up-to-date database of federal and state laws, regulations, and judicial decisions. It had not learned how the law applies to various factual scenarios. In short, calling it an 'AI lawyer' was built on a shaky foundation.

On January 16, 2025, the FTC Commission unanimously approved the final administrative order against DoNotPay, 5-0. It contained three requirements.

First, pay $193,000 in monetary relief.

Second, notify consumers who subscribed between 2021 and 2023 about the FTC settlement and the service's limitations.

Third, provide sufficient evidence before advertising that its services perform like a real lawyer.

$193,000. For a company that had received hundreds of millions in venture funding, that may not be a large sum. But the real significance wasn't in the dollar amount. What mattered was that the FTC took issue with the very phrase 'robot lawyer.' A standard was set: you cannot claim AI replaces a licensed profession without scientific evidence.

Chair Lina Khan and Commissioner Melissa Holyoak issued a joint statement, but its content was not a condemnation of DoNotPay. They expressed hope that AI could give consumers cheaper and more convenient access to services.

The problem was never AI technology itself. It was marketing that exaggerated capabilities without verification.

Browder once made a bold offer on Twitter: 'I'll pay $1 million to any lawyer who walks into court wearing AirPods and argues their case following AI instructions.' The statement drew media attention, but it also caught every lawyer's eye. The legal community responded with criticism that this young man had no idea 'how the law actually works.' Some said he was drunk on his own product.

The DoNotPay case left a lesson for the entire legal tech industry. AI clearly has the potential to lower costs and increase access to legal services.

But the word 'lawyer' carries weight. A lawyer is a credentialed professional. If you're going to claim your product mimics that expertise, you need verification to match. Calling something a 'robot lawyer' without verification isn't marketing. It's deception.

(2) Evolv Technologies: False Claims About an AI Weapons Detector

One day in 2022, a student was stabbed with a seven-inch knife at a middle school in Utica, New York.

A state-of-the-art AI security system had been installed at the school entrance. It was the Evolv Express scanner, sold by Evolv Technologies. The machine failed to detect the knife.

Evolv was headquartered in Massachusetts. Their marketing was ambitious: AI-powered security scanners that were faster, more accurate, and more cost-effective than traditional metal detectors. They targeted the school market in particular.

In the United States, where mass shootings are frequent, school safety was a burning issue. Parents wanted technology that would protect their children, and superintendents needed grounds to say 'our school is safe.'

Evolv's advertising copy was appealing. 'Uses AI to distinguish weapons from everyday personal items.' 'Ignores keys and smartphones; pinpoints only guns and knives.' 'Reduces labor costs by 70% compared to traditional metal detectors.' More than 800 schools across 40 states adopted the system. It was installed at professional sports venues and hospitals too. The school market accounted for half of Evolv's business.

But problems kept surfacing in the field. After the Utica school incident, the school raised the scanner's sensitivity. False alarm rates then soared to 50%. Students' lunch boxes, binders, and water bottles were flagged as weapons. In one Illinois school district, laptops alone triggered more than 85,000 false alarms between August 2023 and April 2024. Five knives were found, but the cost was near-total paralysis of school administration.

The FTC's investigation revealed something worse. According to the complaint, the degree to which the Evolv Express scanner detects weapons 'depends heavily on the sensitivity level set by the user, and higher sensitivity produces more false alarms.' The FTC went further, noting that Evolv calling itself a 'weapons detection system' while claiming it is 'not a metal detector' was 'purely a marketing distinction.' In reality, all the scanner detected was metal; it was powerless against non-metallic weapons.

On November 27, 2024, the FTC announced its action against Evolv.

The order had four key components.

First, a prohibition on claiming the product can detect weapons using AI without sufficient supporting evidence.

Second, a prohibition on false claims about the ability to ignore harmless personal items. Third, a prohibition on comparative claims about accuracy, false alarm rates, or processing speed relative to traditional metal detectors.

Fourth, a prohibition on false claims about labor cost savings.

There was an even more significant measure.

The FTC ordered Evolv to give K-12 school customers who signed contracts between April 1, 2022, and June 30, 2023, the right to terminate their contracts without penalty within 60 days.

Evolv's contracts generally bound schools for multiple years. This order opened an escape hatch for them.

Evolv's response was interesting. In a company statement, they emphasized that "the FTC did not challenge the core efficacy of Evolv's products." They said "the focus of the investigation related to the manner of description in certain historical marketing materials over a period of time." In short, the company claimed the technology itself was fine and only the advertising language needed fixing.

But voices from the field told a different story. Security expert Donald Maye said the FTC's findings highlighted "major gaps in Evolv's technology." School safety consultant Kenneth Trump was more blunt: "Some school leaders may have unknowingly created the very outcomes they were trying to prevent. They could find themselves in the difficult position of explaining why they purchased technology that did not deliver what was promised or implied to them and their school communities."

There is a paradox here. As of January 2025, after the FTC settlement, Evolv announced that not a single customer had requested to terminate their contract. Some schools actually said they were satisfied with the Evolv system. A safety officer from one school district said, "That report means nothing to us. We work with Evolv every day." Opinions on the technology remain divided.

The Evolv case raises important questions. What responsibilities should come with attaching the term "AI" to technology directly tied to safety? If a system that parents trusted with their children's safety fails to perform as advertised, who bears responsibility? Between transparently disclosing a technology's limitations and maximizing marketing impact, what choice should a company make? The answers to these questions remain unfinished.

(3) IntelliVision: False Claims of Facial Recognition Accuracy

IntelliVision Technologies Corp., headquartered in San Jose, California, sold facial recognition software used in home security systems and smart home touch panels. Their website featured confident claims.

"Detects faces of all ethnicities without racial bias."

"Trained on datasets of millions of faces worldwide, with absolutely no gender or racial bias."

"Boasts one of the highest accuracy rates on the market."

In facial recognition technology, the word "bias" is a sensitive subject.

In 2018, MIT researchers Joy Buolamwini and Timnit Gebru published their "Gender Shades" study, revealing that major facial recognition systems showed significantly higher error rates for Black women. Since then, racial and gender bias in facial recognition technology has been a subject of constant debate. IntelliVision's "no bias" claim appeared to be a direct answer to this controversy.

But the FTC's investigation found the opposite.

According to a complaint filed on December 3, 2024, IntelliVision had no evidence to support its claim of "market-leading accuracy." There was also no basis for the assertion that the technology had "absolutely no gender or racial bias." A more serious problem existed. The company advertised that its system was "trained on millions of faces," but in reality it was trained on images of only about 100,000 unique individuals. The company then used techniques to generate variations of those same images.

It helps to understand what "training data" means here. AI models grow by consuming data. For a facial recognition AI to accurately identify faces across different races and genders, it must be trained on facial data representing those diverse races and genders. If training data skews toward a particular race, the AI will recognize faces of that race well but frequently misidentify faces of other races. This is the basic mechanism of "algorithmic bias."

The FTC used publicly available data from the National Institute of Standards and Technology (NIST) to objectively evaluate IntelliVision's algorithm. The results contradicted the company's claims. IntelliVision's algorithm "did not rank among the top 100 highest-performing algorithms." Tests for bias across demographic groups had not been properly conducted either.

The company also made claims about its anti-spoofing technology, asserting that the system could not be fooled by photographs or video images. This too lacked adequate evidence.

On January 8, 2025, the FTC finalized a 20-year administrative order against IntelliVision. The Commission approved it unanimously, 5 to 0. During the public comment period, not a single comment was received. The order had two main requirements.

First, IntelliVision was prohibited from making false claims about the accuracy and efficacy of its facial recognition software, including the comparative performance of its technology across individuals of different genders, ethnicities, and skin tones.

Second, any future claims about the technology's effectiveness, accuracy, or absence of bias would require the company to possess and rely on "competent and reliable testing."

Samuel Levine, Director of the Bureau of Consumer Protection, said in a statement: "If you're going to tout an artificial intelligence system as free from bias, you need to be able to back that up. People who develop and use AI systems are not exempt from basic principles against deceptive advertising."

This was the second major facial recognition case the FTC had brought in the span of a year. In December 2023, the FTC had ordered retailer Rite Aid to stop using facial recognition technology for surveillance purposes for five years, following misuse of the technology. In the Rite Aid case, the issue was not the technology itself but the company's failure to implement reasonable procedures to prevent consumer harm when deploying it.

The IntelliVision case addresses a different dimension of the problem: the gap between a technology's actual performance and its advertised performance. The phrase "no bias" sounds like an ethical claim, but from the FTC's perspective, it is a claim about objective fact. Objective claims require evidence. Asserting "no bias" without evidence is no different from labeling produce "100% organic" while selling conventionally grown crops.

(4) Click Profit / Ascend Ecom: AI Investment Fraud

"AI will make money for you." Is there a more tempting sentence? Click Profit and Ascend Ecom turned this sentence into a weapon.

Their business model was straightforward.

Consumers invest tens of thousands of dollars, and the company opens and operates online stores on platforms like Amazon, Walmart, and TikTok on their behalf. Consumers don't have to do anything. "Cutting-edge AI technology" automatically selects profitable products, manages inventory, and handles shipping.

Thousands of dollars in "passive income" land in your bank account every month. At least, that's how the ads told it.

Click Profit co-founder Craig Emslie appeared frequently in TikTok videos and digital ads.

He said "the stock market, real estate, and precious metals can never offer the level of safety found in a Click Profit investment." The company claimed it had product sourcing partnerships with well-known brands like Nike, Disney, Dell, Colgate, and Marvel. It said it had invested $5 million in developing an AI supercomputer that generated "approximately $100 million in sales."

Consumers poured in their retirement savings and loan proceeds. Joining Click Profit required a management fee of $45,000 to $75,000, plus an additional $10,000 or more for inventory costs. The company took up to 35% of profits generated by customer stores.

The FTC's complaint includes the story of one anonymous consumer. He invested his "life savings" in Click Profit. The result was devastating. He was dropped as a client, with "nothing to show for his payments." When he posted a negative review online, he received contact from Emslie's attorney.

The attorney threatened to sue him and "take everything he and his wife owned." The consumer deleted the review and requested a partial refund. Emslie's response: "Get lost."

Ascend Ecom used similar tactics. The company was run by William Basta and Kenneth Leung and operated under various names since 2021.

Ascend Ecom, Ascend Ecommerce, Ascend CapVentures, ACV Partners, and others. They promised that "cutting-edge" AI tools would enable consumers to quickly earn thousands of dollars in monthly income. They claimed to "maximize client business success with proprietary software and artificial intelligence."

According to the FTC, consumer harm from Click Profit amounted to at least $14 million. The estimated damage from Ascend Ecom was at least $25 million.

Amazon suspended or shut down approximately 95% of the stores Click Profit created for violating platform seller policies. After accounting for Amazon fees, more than 20% of Click Profit stores generated no profit at all, and one-third had total sales below $2,500.

In September 2024, the FTC filed suit against Ascend Ecom. In March 2025, a lawsuit against Click Profit followed. Federal courts froze the companies' assets and halted their operations.

In June 2025, Ascend Ecom's operators agreed to a permanent ban from selling business opportunities and to surrender their assets to the FTC. In August 2025, Click Profit's operators met the same fate. The monetary judgment against Craig Emslie, Patrick McGeoghean, and others totaled $13.6 million. A separate judgment of $7.3 million was entered against Jason Masri. These cases reveal a pattern. AI was the bait. The three letters "artificial intelligence" were used as a tool to boost investor confidence. The FTC's conclusion was that "the heavily promoted AI technology and brand partnerships did not exist, and the promised profits were never realized."

This is not a new phenomenon. During the dot-com bubble of the 1990s, simply adding ".com" to a company's name was enough to attract investment. In 2024, "AI" serves the same function. The more complex a technology, the more people want to believe rather than verify.

Some exploit this tendency. FTC Bureau Director Christopher Mufarrige said in a statement: "Click Profit misled consumers by falsely promising guaranteed passive income using cutting-edge AI technology and exclusive brand partnerships."

C. AI That Wrote Fake Reviews (Rytr LLC Case)

(1) AI-Generated Reviews and Consumer Protection

When we shop online, we look at reviews. Star ratings and comments left by other buyers influence our purchasing decisions.

"This product is really great. I've used it for a week and can already feel the difference." Sentences like this build trust. Reviews serve as a kind of "word-of-mouth map" for unfamiliar products. We trust this map and let it guide our choices.

But what if that map has been fabricated.

Rytr LLC was an AI writing tool that automatically generated various texts when users entered keywords. Blog posts, emails, marketing copy, social media posts. Among its features was one called "Testimonial & Review." Users entered a product name, a star rating, and a few keywords, and the AI produced vivid reviews that read as if written by actual buyers.

Type "hair loss shampoo, 5 stars" and you'd get something like this.

"I've been using it for a week and I can already feel less hair falling out! Highly recommend!" Of course, the AI had never used that shampoo. The AI doesn't have hair. The review was pure fiction.

According to the FTC's September 2024 complaint, the reviews generated by Rytr's service contained "specific and often material details unrelated to users' inputs." These reviews were "almost certainly false for the users who copy and post them online." The FTC alleged that some of Rytr's subscribers used the service to produce "hundreds and, in some cases, tens of thousands of reviews containing potentially false information."

This is where the legal term "means and instrumentalities" enters the picture. An everyday analogy helps explain it. The person who fires the gun isn't the only one responsible. The person who manufactured and sold the gun, knowing it would be used in a crime, can also bear liability. The FTC determined that even though Rytr did not directly post fake reviews, providing a tool that made it easy for consumers to commit fraud was itself unlawful.

The FTC's logic went further. The complaint argued that Rytr's review-generation service had "no or negligible legitimate uses" and that "its sole purpose is to facilitate subscribers posting fake reviews to deceive consumers."

In December 2024, the FTC finalized its order against Rytr. The order had two main elements.

First, a prohibition on engaging in similar unlawful conduct in the future.

Second, a 20-year ban on advertising, promoting, marketing, or selling any service dedicated to, or promoted as, generating consumer reviews or testimonials.

This settlement set off alarm bells for generative AI developers.

"If the AI you built is designed in a way that makes it convenient for wrongdoing, you're an accomplice." The shield of technological "neutrality" appeared to have been pierced. A text-generation AI is, at its core, a tool for producing plausible sentences. When used for fiction writing, it becomes art; when used for fake reviews, it becomes fraud. Can developers really control every possible use.

But this case did not end there. A fierce debate erupted inside the FTC.

(2) Analysis of FTC Commissioner Opinions and Future Direction

In September 2024, the FTC vote approving the complaint and proposed consent order against Rytr was 3 to 2. Commissioners Melissa Holyoak and Andrew Ferguson cast dissenting votes. Both were Republican-appointed commissioners.

Commissioner Ferguson's dissent was scathing.

He pointed out that "Rytr's tool can be used for both lawful and unlawful purposes." It was not an "inherently deceptive" tool, he argued. He went further, emphasizing that "there is no allegation that Rytr knew its tool was being used deceptively."

His central argument was this: "Categorically treating generative AI tools as illegal simply because they could potentially be used for fraud is contrary to our precedent and common sense. And it threatens to turn honest innovators into lawbreakers and risks strangling a potentially revolutionary technology in its cradle."

Commissioner Holyoak struck a similar tone: "Today's misguided complaint and misapplication of Section 5 is likely to chill innovation in the AI space."

These dissents were not mere legal disagreements. They revealed a philosophical fork in AI regulation. One side holds that "the risk of potential harm and market corruption alone is sufficient to justify action." The other holds that "punishing the tool itself without evidence of actual harm will stifle innovation."

This debate met a dramatic reversal one year later.

On December 22, 2025, the FTC voted to withdraw the final consent order against Rytr. It was the first case implementing the Trump administration's AI Action Plan. Ferguson, now Chair, had turned the logic from his own dissent into policy.

In the new FTC's announcement, Bureau Director Christopher Mufarrige stated: "Simply condemning a technology or service because it has the potential to be problematic is inconsistent with law and order." The FTC concluded that "the specific facts alleged in the complaint do not support a Section 5 violation." It could not "imagine a need to maintain an order that is not based on a cognizable violation of the law and therefore does not protect consumers."

The Rytr reversal carries significant implications for the future direction of U.S. AI regulation.

First, under the Trump administration, the FTC is likely to prioritize promoting AI innovation and refrain from taking enforcement action against AI tool providers without clear evidence of harm and a solid legal basis.

Second, application of the "means and instrumentalities" doctrine will become more cautious, with a strengthened position that the mere potential for misuse of a tool is insufficient grounds for enforcement.

Third, AI-related enforcement actions are expected to move toward requiring concrete evidence of actual consumer harm and deception.

Yet this does not mean a free-for-all. Even after the Rytr withdrawal, the FTC has continued imposing strong sanctions against blatant fraud operations like Click Profit and Ascend Ecom. The tip of the regulatory sword has simply shifted from "technology developers" to "malicious users" and "deceptive marketers."

The Rytr case symbolizes the dilemma of AI regulation. Should we regulate the tool itself to prevent fraud, or should we allow the tool for the sake of innovation and punish only misuse. The answer to that question can change depending on who occupies the White House. The same facts under the same statutory provisions can lead to entirely opposite conclusions.

For companies, this means uncertainty. What is legal today may be illegal tomorrow, and what is illegal today may be legal tomorrow. But some things remain certain. Exaggerating the capabilities of AI technology, or using AI as a pretext to deceive consumers and take their money, will not be tolerated under any administration. The core message of Operation AI Comply stands: "Slapping the word AI on something doesn't turn fraud into innovation." In the next chapter, we turn our gaze across the Atlantic. While the United States was focused on catching fraudsters, the European Union was aiming at something far larger. Armed with the GDPR and the EU AI Act, European regulators have been waging a different kind of war against AI companies, targeting personal data and human rights.