Recent regulatory activity underscores the critical need for companies to take "AI-washing" seriously, according to Andrew Lustigman and Barry Greenbaum of Olshan. The core principle remains: a company’s marketing materials should not be solely substantiated by its vendors’ claims. As artificial intelligence rapidly integrates into products and services across industries, a growing chasm between marketing language and technological reality is emerging, leading to increased scrutiny from regulators like the Federal Trade Commission (FTC) and advertising industry bodies.
The proliferation of AI-related claims in corporate communications is undeniable. Companies are increasingly touting "AI-powered" personalization, "AI-driven" recommendations, automated customer service solutions, and predictive analytics. These assertions are disseminated across a wide array of platforms, including company websites, sales presentations, responses to Requests for Proposals (RFPs), customer contracts, vendor disclosures, and executive interviews. However, in many instances, the underlying technology plays a minimal, or even non-existent, role in delivering the advertised benefits. This discrepancy, termed "AI-washing," draws parallels to the well-established issue of "greenwashing" related to environmental claims and is now attracting significant attention from enforcement agencies.
The requirement for advertising substantiation is not a new concept. Regulators have long mandated that companies possess a reasonable basis for objective claims before they are made public. What is novel in the current landscape is the inherent complexity of AI claims and the formidable challenge of evaluating the intricate technology that underpins them. Despite this complexity, the fundamental standard for substantiation has not changed: companies are still obligated to have a reasonable basis for their objective assertions. The primary difficulty lies in the evolving nature of AI, which makes establishing, evaluating, and consistently maintaining this basis significantly more arduous.
Defining the AI Claim: A Crucial First Step
A primary hurdle in addressing AI-washing lies in the ambiguity of the claims themselves. Terms such as "AI-powered," "AI-driven," "intelligent," and "automated" can carry vastly different meanings depending on the specific context. For instance, a consumer might reasonably interpret "AI-powered customer service" to mean that an AI system handles interactions with minimal or no human intervention. In reality, the product might only utilize AI to suggest responses, with human oversight still required before any communication is sent.
A notable case illustrating this challenge involved Samsung, which faced scrutiny from the National Advertising Division (NAD), the advertising industry’s self-regulatory body. The NAD investigated claims made for Samsung’s Bespoke refrigerators, specifically those suggesting that their "smart" connectivity was AI-driven. Following the NAD’s review, Samsung agreed to discontinue these specific claims, highlighting the need for precision in articulating the role of AI.
Before embarking on the process of substantiating an AI claim, legal and compliance teams must engage in a critical self-assessment. The foundational question they should ask is: "What would a reasonable customer understand this statement to mean?" This inquiry compels organizations to clearly articulate the actual functionality of the technology. Does it make independent decisions, assist human decision-makers, or perform a limited function within a broader, rules-based system? A clear definition of the claim is the indispensable first step in identifying the precise evidence required to support it.
Evidence Must Align with the Promise
Once an AI claim has been clearly defined, the subsequent and equally vital questions revolve around the availability of supporting evidence and, crucially, whether that evidence directly corresponds to the promise made in the claim.
Consider a company asserting that its AI system is "95% accurate." This claim immediately triggers a series of important questions: How was this accuracy measured? What specific datasets were used for testing? Against what benchmarks or conditions was the system evaluated? Similarly, a claim stating that the system is "more accurate than human reviewers" introduces a different set of considerations. This would necessitate identifying the human reviewers involved, the specific tasks they performed, and whether the comparison was statistically meaningful and conducted under comparable conditions. If a product is advertised as "fully automated" but employees routinely review or modify its output, the company must critically assess whether this human involvement materially alters the customer’s understanding of the claim.
Regulators have actively pursued companies making unsubstantiated AI claims. The FTC’s enforcement action against Workado, LLC, serves as a prime example. The FTC challenged the company’s claims that its AI content detector was "98% accurate." Independent testing, however, revealed the accuracy rate to be significantly lower, closer to 53%. Workado settled with the FTC, agreeing to stringent prohibitions against making representations about the effectiveness of its AI products without competent and reliable supporting evidence.
More recently, the FTC took action against Cox Media Group and two smaller marketing agencies concerning their "Active Listening" branded marketing service. The FTC alleged that these companies deceptively claimed their service utilized a special algorithm to listen to consumers’ conversations overheard by smart devices, thereby facilitating targeted advertising in desired locations. The FTC contended that the service did not actually engage in such listening or utilize voice data, and moreover, that consumers had not provided explicit consent for this service. Cox Media Group and the marketing agencies settled these actions, agreeing to pay nearly $1 million and accepting prohibitions on claims related to the qualities or features of their advertising or marketing services, the collection and use of voice data, and geographic targeting capabilities.
In the realm of self-regulation, Horizon Brands faced a challenge from the NAD regarding advertising claims for its Tiny Traveler "AI-powered Smart Baby Monitor Solution." While the presence of an AI chip confirmed the use of AI technology, the NAD recommended that claims related to AI "emotion detection" and "motion detection" acknowledge their limitations. Furthermore, the NAD advised discontinuing claims that the monitor could ensure infant safety, indicating that even with AI present, the scope of its capabilities must be accurately represented. This case underscores the importance of being precise about the availability of features and disclosing any limitations on performance or operating conditions.
Substantiating Through Vendors: A Complex Interplay
The challenge of substantiating AI claims becomes significantly more intricate when a company relies on third-party AI technology. Retailers, financial institutions, software developers, and a multitude of other businesses are increasingly integrating AI tools procured from external vendors. These vendors often make robust claims about their technology, describing it as highly accurate, autonomous, secure, or capable of delivering specific business outcomes. Such representations frequently find their way into the customer company’s own marketing collateral, sales materials, or RFP responses. However, a critical point to remember is that a vendor’s marketing materials should not automatically serve as a company’s sole substantiation record.
Companies must diligently understand precisely what their vendors are supplying and scrutinize the evidence that supports material claims about the underlying technology. Vendor claims must be independently verified before they are incorporated into a company’s own marketing or customer communications. Contracts can play a vital role in facilitating this process by obligating vendors to provide detailed technical documentation, comprehensive testing reports, and timely notification of any material changes to their technology. Nevertheless, contractual protections alone are insufficient as a substitute for a thorough understanding of the product. If a company assures its customers of a specific AI-enabled benefit, simply pointing to a vendor’s website after the fact will likely not suffice to demonstrate that the company’s own statement was adequately supported.
Ensuring Consistency Across the Organization
The widespread adoption of AI often means that claims span multiple departments within an organization, thereby amplifying the risk of inconsistencies. Marketing departments might approve specific language for a company website, while product developers may describe the identical feature in a subtly different manner during a sales presentation. A salesperson might then make an even broader representation to a prospective client. An executive, during an interview, might characterize the technology in yet another distinct way. Individually, each statement might appear innocuous. However, when viewed collectively, these varied descriptions can construct an impression of the product’s capabilities that is considerably more expansive than what the technology can actually deliver.
To mitigate this risk, a coordinated review process is essential. This process should involve legal or compliance teams, product or engineering departments, and the business or marketing teams responsible for customer communications. Such collaboration ensures that all stakeholders operate from a unified understanding of the technology and can clearly identify the evidence supporting any claims. The objective is not to enforce identical language across every communication channel but to guarantee that variations in messaging do not convey materially different or misleading impressions of the technology’s capabilities.
The Imperative of Ongoing Substantiation
Finally, companies must establish and maintain a comprehensive and current record for all material AI claims. This record should meticulously identify:
- The specific claim being made: Precisely what is being asserted about the AI’s functionality or benefit.
- The evidence supporting the claim: This includes documentation, testing results, internal evaluations, and any other data that substantiates the assertion.
- The date the claim was first made and last reviewed: This helps track the currency and relevance of the substantiation.
- The individuals or departments responsible for the claim and its substantiation: Clearly defined ownership is crucial for accountability.
- Any limitations or caveats associated with the claim: Transparency about the boundaries of the AI’s performance is vital.
Maintaining a live, periodically updated record ensures the ongoing accuracy and defensibility of AI claims. Before approving any AI claim, companies should undertake a thorough review. This involves scrutinizing what is being communicated, considering what a reasonable customer would understand from that communication, and assessing the existing evidence. Organizations that consistently and rigorously answer these questions each time an AI claim is formulated will be best positioned to build lasting trust with their customers, effectively navigate the evolving regulatory landscape, and maintain their credibility in the market. The proactive and diligent management of AI claims is no longer just a best practice; it is an essential component of responsible business conduct in the age of artificial intelligence.
