Artificial intelligence has become one of the most powerful marketing terms in business. Companies now describe ordinary software, automation tools, analytics dashboards, chatbots, screening systems, investment platforms, and customer-service products as “AI-powered,” “machine-learning enabled,” “algorithmic,” “predictive,” or “autonomous.” In many cases, those descriptions are fair. Businesses are legitimately using AI to improve speed, accuracy, personalization, and scale. The legal problem begins when the marketing races ahead of the product. When a company sells a feature as “powered by AI” before the feature actually works as represented, or before the company has built the technology it claims to have, that statement can become more than sales language. It can become evidence in a fraud case, a breach-of-contract dispute, a warranty claim, a consumer-protection action, an SEC enforcement matter, or a securities class action. ¹
The phrase “AI washing” describes the practice of overstating, mischaracterizing, or fabricating the role artificial intelligence plays in a product, service, or business model. It is the technology-law version of “greenwashing.” The core issue is not whether a company uses the letters “AI.” The issue is whether the company’s public and private statements create a materially false impression about what the technology does, how it works, how mature it is, whether it has been tested, whether it is actually deployed, whether customers are receiving the promised benefit, and whether revenue or business prospects depend on those claims. A company may say that it is “exploring AI,” “developing AI tools,” or “using automation to support internal workflows” without creating the same risk as a company that claims its product already uses proprietary AI to deliver a specific measurable result. The more concrete the claim, the more likely it is to be tested against reality.
The legal exposure is especially acute for businesses that market AI features before the features work as advertised. A product roadmap, prototype, limited beta, manual workaround, outsourced data-labeling process, or traditional rules-based workflow may be useful internally, but it cannot safely be marketed as a fully functioning AI capability unless that is true. The distinction matters because customers and investors often pay a premium for AI. Customers may choose one vendor over another because the vendor claims that its AI reduces labor costs, detects risks, predicts outcomes, improves safety, automates decisions, personalizes services, or unlocks proprietary insights. Investors may assign a higher valuation because management says that AI is driving growth, retention, margins, or competitive advantage. When those statements later prove inaccurate, the resulting litigation rarely turns on whether AI is exciting. It turns on whether the defendant said something specific, whether the statement was false or misleading when made, whether others relied on it, and whether losses followed.
The SEC’s 2024 AI-washing enforcement actions against Delphia and Global Predictions illustrate how quickly marketing language can become enforcement evidence. The SEC alleged that Delphia made false and misleading statements in filings, advertisements, and online materials about its use of artificial intelligence and machine learning in its investment process. According to the SEC’s order, Delphia had stated that client data would be used to make its AI smarter and support investment decisions, but the company had not developed or used those capabilities as claimed. Global Predictions faced similar scrutiny for claims about “AI-driven forecasts,” its status as an AI financial adviser, performance-related assertions, and other marketing statements. The firms resolved the matters without admitting or denying the SEC’s findings and agreed to civil penalties. The message was not that investment advisers cannot use AI in marketing. The message was that AI claims must be accurate, substantiated, and consistent with what the company actually does. ¹
That enforcement theory has implications far beyond investment advisers. The SEC’s public statements warned that public issuers making AI-related claims must remain vigilant when those statements may be material to investors. A public company does not need to be an AI company to create securities exposure. A manufacturer, retailer, software provider, healthcare company, financial-services business, or security-technology company can face risk if it tells investors that AI is improving performance, reducing costs, generating demand, or differentiating the company in the market, while internal documents show that the technology is immature, manual, unreliable, outsourced, delayed, or not materially contributing to results. In securities litigation, the marketing claim and the investor claim often merge. A statement made to customers may also appear in investor presentations, earnings calls, SEC filings, press releases, or analyst communications. Once a company connects AI to revenue, margins, growth, or market leadership, it has moved from product puffery into potentially material disclosure territory.
Securities class-action filings and commentary show that AI-related litigation risk is not theoretical. Recent analyses report that AI-related securities complaints have increased in recent years and that plaintiffs are focusing on recurring themes: alleged exaggeration of AI capabilities, alleged concealment of manual or non-AI processes, allegedly misleading statements about AI-driven revenue or demand, and alleged failures to disclose limitations, risks, or defects. Some complaints allege that companies described AI tools as proprietary or advanced when the technology was less sophisticated than advertised. Others allege that companies attributed growth to AI when demand was driven by other factors, or that AI initiatives were cannibalizing revenue rather than accelerating it. The important litigation trend is not that every AI statement produces liability. It is that plaintiffs are learning how to plead AI statements using familiar securities-law theories. AI becomes the factual setting; the claim remains a traditional allegation that the company misled the market. ²
The Federal Trade Commission’s action involving Evolv Technologies shows the same concept in the product-marketing context. The FTC alleged that Evolv made deceptive claims about AI-powered security screening systems, including claims about detecting weapons, ignoring harmless personal items, improving accuracy and speed, reducing false alarms, and lowering labor costs. The FTC’s allegations were particularly significant because the marketed product was used in schools and safety-sensitive environments. In that setting, an AI claim is not merely a branding flourish. It can influence procurement decisions, public-safety judgments, staffing levels, and contract commitments. The FTC’s theory underscores a broader rule for businesses: performance claims about AI must be backed by evidence before they are used in sales materials. A company should not wait until litigation to determine whether its testing supports its marketing. ³
For private businesses, the most immediate risk often begins with the sales process. AI claims are commonly made in pitch decks, proposals, capability statements, demos, webinars, website copy, emails, responses to requests for proposals, and statements of work. Those materials can later be used to show what the customer was promised. If the customer buys a product because it was described as AI-enabled, predictive, automated, or self-learning, and the delivered system does not perform that way, the customer may frame the dispute as breach of contract, breach of warranty, fraudulent inducement, negligent misrepresentation, or violation of a consumer-protection statute. Even when the final contract contains integration clauses, limitation-of-liability provisions, warranty disclaimers, or acceptance procedures, the pre-contract representations may still shape the litigation. They can influence how a court views reliance, inducement, materiality, contract interpretation, and the commercial reasonableness of the parties’ conduct.
The line between aggressive marketing and actionable misrepresentation often depends on specificity. A vague statement that a company is “innovative” or “using cutting-edge technology” is less likely to support a claim standing alone. A statement that the product “uses proprietary AI to reduce invoice-processing time by 70%,” “detects threats with greater accuracy than traditional systems,” “uses customer data to train predictive investment models,” “automatically identifies defects in real time,” or “replaces manual review” is different. Those statements describe present capabilities or measurable outcomes. They invite proof. If the product does not use AI in the represented manner, the claim is based on a limited lab test rather than real-world deployment, if the result depends on heavy human intervention, or if the model fails under ordinary use conditions, the claim may become legally actionable.
Timing is also critical. Businesses often defend AI statements by arguing that the product was still under development or that the claim reflected a future goal. That distinction can help, but only if the marketing actually made the timeline clear. There is a legal difference between saying “we are developing an AI feature that we expect to launch next year” and saying, “our platform uses AI to optimize your operations today.” There is also a difference between a beta program that is disclosed as experimental and a commercial deployment sold as proven. A company that markets a not-yet-working AI feature as a current capability may create fraudulent-inducement risk because the customer’s decision to contract is based on a present factual representation. The problem is not that the company hoped to build the feature. The problem is that it allegedly sold the hope as if it already existed.
Michigan law provides a useful framework for understanding this risk. Michigan courts have long required a fraud plaintiff to prove a material representation, falsity, knowledge or reckless disregard of falsity, intent that the representation be acted upon, actual reliance, and resulting injury. Michigan courts also distinguish between statements of existing fact and promises of future conduct. A broken promise is usually a contract issue, not fraud, unless the plaintiff can show that the promise was made in bad faith and without intent to perform at the time it was made. In the AI-washing context, that distinction matters. A vendor’s missed development deadline may support a contract claim, but a vendor’s statement that an AI feature already existed, already used particular data, or already achieved particular results may support a fraud theory if the statement was false when made.⁵
Michigan contract law also gives customers a straightforward path when the AI claim becomes part of the bargain. A breach-of-contract claim generally requires proof of a contract, breach, and damages. If the contract, statement of work, purchase order, service description, warranty, or incorporated proposal promises a specific AI capability, the failure to deliver that capability may be pleaded as breach. The same analysis may apply where the vendor promises real-time automation, machine-learning recommendations, predictive analytics, or autonomous decision-making and instead delivers a conventional dashboard, a manual workflow, or an unreliable tool that does not meet specifications. In that setting, the claim may not require proving that the vendor intended to deceive anyone. The customer may simply argue that it paid for one thing and received another.⁵
The contract-versus-fraud boundary is important because many technology disputes involve both theories. Michigan’s economic-loss doctrine can restrict tort claims when the alleged injury is purely economic, and the dispute is fundamentally about disappointed contractual expectations. But Michigan also recognizes that fraud in the inducement may be different when the alleged misrepresentation is extraneous to the contract and caused the plaintiff to enter the agreement in the first place. This distinction is especially relevant to AI marketing. If the alleged misrepresentation merely restates the same performance obligation contained in the contract, the dispute may sound primarily in contract. If the alleged misrepresentation concerns pre-contract deception about existing AI capabilities, proprietary technology, testing, data usage, or readiness for deployment, a plaintiff may argue that it was fraudulently induced to contract on false premises.⁵
Businesses should therefore treat AI descriptions as controlled legal statements, not casual marketing adjectives. The review process should begin by asking what the product actually does today. Does the feature use machine learning, generative AI, rules-based automation, third-party AI, human review, or a combination? Is the AI used in production or only in development? Is it customer-facing or internal? Does the model make decisions, generate recommendations, rank options, classify data, summarize documents, detect anomalies, or merely assist employees? Does the company have test results supporting claims about accuracy, speed, cost savings, or reliability? If the answer is uncertain, the marketing should be revised before publication. The safest language is usually precise language that describes the actual function, the deployment status, and any limitations.
Substantiation is the practical dividing line between responsible AI marketing and AI washing. A company that claims its AI improves accuracy should know compared to what baseline, in what environment, over what sample size, during what testing period, and subject to what limitations. A company that claims its AI reduces labor costs should know whether the claim reflects actual customer results, internal estimates, pilot data, or assumptions. A company that claims its AI is proprietary should know which components are proprietary and which depend on third-party models, open-source tools, licensed systems, or human labor. A company that claims its AI “learns” from customer data should know whether customer data is actually used for training, fine-tuning, retrieval, ranking, personalization, or none of those things. Vague enthusiasm is not a substitute for evidence.
Internal inconsistency is another major litigation trigger. Companies often say different things to different audiences. Product teams know that a feature is in beta. Sales teams describe it as available. Executives tell investors that AI is driving growth. Customer-success teams quietly rely on manual workarounds. Compliance teams add risk disclosures saying the technology may not work as expected. Plaintiffs’ lawyers look for those gaps. Regulators do too. In litigation, emails, Slack messages, Jira tickets, product roadmaps, model evaluations, customer complaints, support logs, renewal notes, and board materials may all be compared against public claims. A company that wants to avoid AI-washing exposure should align its internal truth with its external story.
Disclaimers help, but they do not cure falsehoods. A well-drafted contract can define the product, limit warranties, exclude reliance on extra-contractual statements, cap damages, and state that future features are not guaranteed unless expressly included. Those provisions are important. But a disclaimer is not a license to make false statements in marketing. If a vendor says in a sales deck that an AI feature is already operational, then buries a limitation in a contract that the customer may not receive until later, the vendor should not assume the disclaimer will solve the problem. Similarly, public companies cannot neutralize concrete AI claims with generic risk factors that fail to disclose known limitations. Risk disclosure must be tailored to the actual risk; it should not describe as hypothetical a problem the company already knows exists.
Companies should be especially careful when AI claims involve safety, finance, healthcare, employment, education, insurance, credit, security, or other high-impact settings. In those areas, customers and end users may rely on AI claims in ways that affect physical safety, money, access, opportunity, or legal rights. Claims about accuracy, bias reduction, fraud detection, threat detection, medical triage, investment performance, hiring efficiency, or compliance automation can invite heightened scrutiny. The more consequential the use case, the more important it is that the company possess competent and reliable evidence before making the claim. Regulators are less likely to view unsupported AI performance statements as harmless puffery when the product affects children, patients, investors, employees, or public safety.
AI-washing risk is not limited to statements that the technology exists. It also includes statements about what the technology replaces. Many companies market AI as a way to eliminate manual review, reduce staffing, automate judgment, or remove human error. If human review remains essential, that fact may need to be disclosed, especially if customers are buying the product to reduce labor or increase speed. A business may truthfully say that AI assists human reviewers, triages tasks, or drafts recommendations. It should not say that the system makes autonomous decisions if employees are actually making the decisions. It should not claim that AI eliminates the need for manual review if manual review is necessary for ordinary performance. The law does not prohibit hybrid human-AI systems. It prohibits misleading descriptions of them.
The same principle applies to third-party AI. Many businesses build valuable products using external models, APIs, licensed platforms, and open-source systems. There is nothing inherently wrong with that. The risk arises when a company markets those capabilities as proprietary, exclusive, internally developed, or uniquely trained when they are not. A customer may care whether a vendor owns the model, controls the training data, can explain the outputs, can maintain the system if a third-party provider changes terms, or can meet confidentiality and data-security obligations. An investor may care whether the company has defensible AI assets or is merely wrapping a common model in a branded interface. Precision avoids later accusations that the company sold borrowed technology as proprietary intelligence.
For public companies, the disclosure issue should be managed at the governance level. AI-related statements should not be left solely to marketing teams or investor-relations personnel. Legal, compliance, finance, product, engineering, and data-science personnel should review material claims before they appear in earnings scripts, investor decks, annual reports, offering materials, websites, or press releases. The company should maintain records showing the basis for claims about AI capabilities, performance, revenue impact, customer adoption, and risk mitigation. If a claim depends on limited pilot data, that limitation should be clear. If AI is not yet material to revenue, the company should avoid implying that it is. If there are known failures, customer complaints, implementation delays, or high human-intervention rates, the company should consider whether existing disclosures remain accurate.
Private companies should adopt the same discipline, even if they are not SEC registrants. AI claims can affect venture financing, lender diligence, mergers and acquisitions, customer contracts, government procurement, and employment recruiting. A startup that exaggerates its AI capabilities in a pitch deck may face investor claims if funding is obtained on false premises. A vendor that exaggerates AI performance in an RFP response may face procurement remedies or contract termination. A company that sells AI-enabled compliance software may face indemnity demands when the product fails to perform. The business may also create reputational damage that exceeds the immediate legal exposure. In AI markets, trust is a commercial asset.
The safest approach is not to avoid talking about AI. It is to talk about AI accurately. Businesses can and should explain how they use AI when the explanation is true, specific, and supportable. They can describe current functionality, planned features, beta limitations, human oversight, third-party dependencies, testing methodology, and appropriate use cases. They can use carefully qualified language when a product is developing. They can distinguish between “AI-assisted,” “AI-generated,” “AI-enhanced,” “rules-based automation,” and “machine-learning model” where those distinctions matter. They can avoid absolute claims such as “always,” “all,” “guaranteed,” “fully autonomous,” or “eliminates” unless the evidence justifies them. In many cases, a more modest claim is not only safer; it is more credible.
AI-washing lawsuits will likely continue because AI remains commercially valuable and technically difficult for outsiders to evaluate. Customers and investors cannot easily inspect a model, training pipeline, data architecture, or performance benchmark. They often must rely on what the company says. That asymmetry creates temptation for sellers and suspicion for buyers. Courts and regulators will therefore focus on ordinary legal questions: What exactly was said? Who said it? When was it said? What did the company know? Was the statement material? Was it relied upon? Was there evidence supporting it? Did the truth later emerge in a way that caused loss? AI may be new, but the litigation framework is familiar.
For businesses, the lesson is straightforward. “Powered by AI” should be treated as a factual claim unless it is clearly framed as aspiration or general branding. Before a company says that its product uses AI, the company should be prepared to explain what that means. Before it says that AI improves performance, it should have data. Before it tells investors that AI is driving growth, it should understand the actual revenue contribution. Before it tells customers that AI replaces manual work, it should know whether manual work is still required. And before it signs a contract based on AI capabilities, it should make sure the contract accurately states what is being delivered. The legal risk is not that a company markets innovation. The risk is that it markets unfinished or unsupported technology as if it already works. When that happens, “powered by AI” can become the first exhibit in a lawsuit.
Contact Tishkoff
Tishkoff PLC specializes in business law and litigation. For inquiries, contact us at www.tish.law/contact/. & check out Tishkoff PLC’s Website (www.Tish.Law/), eBooks (www.Tish.Law/e-books), Blogs (www.Tish.Law/blog) and References (www.Tish.Law/resources).
Footnoted Sources
1- U.S. Securities and Exchange Commission, “SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence,” Press Release No. 2024-36, March 18, 2024; In the Matter of Delphia (USA) Inc., Investment Advisers Act Release No. 6573, Administrative Proceeding File No. 3-21894, March 18, 2024; In the Matter of Global Predictions, Inc., Investment Advisers Act Release No. 6574, Administrative Proceeding File No. 3-21895, March 18, 2024. https://www.sec.gov/newsroom/press-releases/2024-36
2- Melanie E. Walker and Emma Peplow, “AI-related securities class action filings are on the rise: Key observations,” DLA Piper, September 2025; Cornerstone Research and Stanford Law School Securities Class Action Clearinghouse, “Securities Class Action Filings—2025 Midyear Assessment,” summarized in Cornerstone Research press release dated July 30, 2025. https://www.dlapiper.com/en-us/insights/publications/2025/09/ai-related-securities-class-action-filings-are-on-the-rise-key-observations
3-Federal Trade Commission, “FTC Takes Action Against Evolv Technologies for Deceiving Users About its AI-Powered Security Screening Systems,” November 26, 2024. https://www.ftc.gov/news-events/news/press-releases/2024/11/ftc-takes-action-against-evolv-technologies-deceiving-users-about-its-ai-powered-security-screening
4- Stanford Law School Securities Class Action Clearinghouse, “Current Trends and Related Filings,” Artificial Intelligence category, noting AI-related securities class-action filings and related trend data. https://securities.stanford.edu/
5- Hi-Way Motor Co. v. International Harvester Co., 398 Mich. 330; 247 N.W.2d 813 (1976); Huron Tool and Engineering Co. v. Precision Consulting Services, Inc., 209 Mich. App. 365; 532 N.W.2d 541 (1995); Bank of America, N.A. v. First American Title Insurance Co., 499 Mich. 74; 878 N.W.2d 816 (2016). https://law.justia.com/cases/michigan/supreme-court/1976/56828-2.html