How to spot AI washing in IT tools

On March 18, 2024, the Securities and Exchange Commission (SEC) fined two investment advisers for making false and misleading statements about their use of AI to help clients. Both companies agreed to pay a penalty. Delphia (USA) paid $225,000 and Global Predictions Inc. paid $175,000. In the SEC announcement, then-Chair Gary Gensler used a term for the practice: AI washing.

Since those were finance companies, the investors had regulators checking the claims. But many IT buyers don’t have that safeguard. The IT industry is full of vendors labeling products as “AI-powered,” “agentic,” or “autonomous.” Some of these products have real AI capabilities, but others are existing products relabeled to meet current demand.

Unfortunately, buyers are the ones who take the hit when a product is incorrectly labeled. We’ll look at what AI washing is, how the term was established, and how you can spot it before signing with a vendor.

What AI washing actually means

AI washing refers to companies making inflated claims regarding their AI use. Gensler first mentioned AI washing in a December 2023 speech, where he compared it to greenwashing, a term that describes companies overstating their environmental friendliness. Gensler said that, just as companies shouldn’t greenwash, they also shouldn’t AI wash.

In March 2024, the SEC applied the term in its actions against Delphia and Global Predictions. According to the agency’s order, from 2019 to 2023, Delphia told clients that its investment advice was guided by machine learning. In July 2021, Delphia admitted to the SEC that it didn’t incorporate client data into its machine learning models as advertised, but it continued to make those claims in emails and press releases for the next two years.

Global Predictions used messaging that indicated it was the “first regulated AI financial advisor,” and said it offered “expert AI-driven forecasts.” The SEC found those statements misleading.

While both companies agreed to pay the penalties, neither admitted nor denied the SEC’s findings. The SEC took action under existing regulations about deceiving investors rather than setting new rules specific to AI claims. That’s the key lesson from both cases. When a business makes an AI claim, it’s held to the same standard as other factual claims, and that claim could eventually invite scrutiny.

Why IT buyers are exposed to the same pattern

The tech industry has its own AI washing case. In 2025, New York federal prosecutors indicted Albert Saniger, who founded a shopping app called Nate, on securities and wire fraud charges. According to the indictment, Saniger told investors that built-in AI handled the shopping for customers when in reality, the company was outsourcing the work to overseas contractors. Prosecutors allege that the app used almost no automation. Saniger has pleaded not guilty.

If proven, the allegations in the Saniger case showcase an issue that IT buyers recognize all too well. An app claims autonomy when in actuality, humans are handling the workload behind the scenes.

The issue is more widespread than many IT teams realize. A June 2025 report from Gartner estimated that only about 130 of the thousands of vendors marketing agentic AI capabilities actually offer them. Gartner referred to the practice as “agent washing,” which it defined as the rebranding of assistants, robotic process automation, and chatbots as agents. Gartner also predicted that more than 40% of agentic AI projects will be canceled by the end of 2027.

We covered agent washing in an article titled Are AI agents actually autonomous? AI washing is a broader issue. With AI washing, businesses slap an “AI-powered” label on an existing product, along with claimed generative AI features and performance numbers that can’t be verified.

But increasing scrutiny could bring problems for vendors who mislabel their products. In an analysis from the Harvard Law School Forum on Corporate Governance, attorneys cautioned that testing AI claims is becoming easier, leaving companies vulnerable to lawsuits and regulatory action. Although that warning was specific to public-company disclosures, buyers can also test AI claims before spending money on an AI-labeled product.

Four patterns to watch for

In IT procurement, AI washing tends to follow a few distinct patterns. Sometimes it’s unintentional and results from marketing teams stretching a feature description. In every case, though, the buyer is the one who pays for it.

A rebrand with no underlying change

Last year, a vendor’s product was marketed as “automation” or “scripted workflows.” This year, the same product has been relabeled as an AI agent. The release notes may mention AI, but nothing seems to have changed. A help desk that routes tickets using keyword rules isn’t agentic simply because the vendor added an AI badge. Ask what changes were made that justified the new label.

A demo that can’t survive production

Demos tend to be carefully curated, with predictable queries and an expert standing by. Production rarely runs that smoothly. A security solution may handle a staged alert flawlessly during a demo, only to need manual oversight once it’s live. At that point, you’ve paid for an agent but ended up with an assistant that still needs a human at the keyboard.

An unfalsifiable benchmark

If you’ve shopped AI vendors lately, you’ve likely seen claims like “resolves 80% of issues” and “saves 10 hours per technician per week.” Numbers like these are impressive, but without a way to verify them, they’re meaningless. Eighty percent of which issues, over what timeframe, in what environment? Delphia shows how long an unverifiable claim can survive. Clients had no way to check whether their data fed the models, and the claims stayed in emails and press releases for two years after the company admitted to the SEC that they weren’t accurate.

Vendors often access AI capabilities through third-party APIs, and that’s a perfectly legitimate way to build software. The problem comes when a vendor misrepresents a third-party integration as its own proprietary software.As you’re shopping vendors, ask which parts are proprietary and which come from third parties, and how your data will be protected. Clear answers to both separate convincing AI agents from trustworthy ones.

What a falsifiable claim looks like

A falsifiable claim is one a buyer can test and potentially disprove. Vendors that make falsifiable claims are specific and verifiable in their wording, and they let the buyer test the claim in their own environment before committing.

One example of that structure is Atera, an IT management platform with 13K customers across 120+ countries. Robin by Atera is an AI technician that remediates Tier-1 and complex Tier-2 technical incidents end to end by autonomously taking real actions on devices, servers, mainframes, and networks, without needing a technician in the loop.

Atera reports up to 92% autonomous resolution of Tier-1 and complex Tier-2 technical incidents across Robin accounts, as of May 2026. That’s the type of statement customers should scrutinize, so here is what sits behind it: the scope is Tier-1 and complex Tier-2 technical incidents, the basis is Robin accounts, and the figure is dated. 

When considering an autonomous help desk solution, IT teams should ask how tickets will be classified in their environment, since that defines which tickets count toward the stated success rate.

Five-question checklist for your next vendor call

When meeting with a vendor pitching AI capabilities, ask the following questions.

1. What specific work processes does this handle on its own? Ask for a list of tasks it can complete end to end without human intervention.
2. How are your success rates measured? Resolution and accuracy rates mean little unless the vendor can explain which activities were counted and which were excluded in the testing that produced that number.
3. What happens when something goes wrong? Who is accountable? Look for information on monitoring, escalation paths, rollback options, and contract language that assigns responsibility for a system’s failure to perform as advertised.
4. Can we test the product in our own environment before committing? Testing a product live using your own data beats a carefully produced demo every time.
5. What will you commit to in writing? Ask which performance claims will appear in the contract, such as scope, success rate, and measurement method, and what recourse you have if the product doesn’t deliver. Claims that live in the sales deck but never reach the paper deserve a second look.

To learn more about the terminology behind each of these questions, see our agentic AI glossary. Check out our Smart AI adoption for IT guide for a broader framework that covers security and compliance.

Ask for a claim you can check

Spotting AI washing requires asking probing questions before signing a contract. Many companies live up to their claims, and this will show in their answers. The two SEC cases show a simple standard. Regulators want companies to make AI claims that can be validated, and IT buyers should have the same expectation.

Vendor claims should be detailed, including numbers and scope of work. They should also be testable before you sign, and put in writing where it matters. Without that transparency, a claim is just marketing spin. If you want a look at how this works in practice, drop the hype and put your current vendor through Atera’s five-question Autonomous IT test.