Of the 9,214 AI funding rounds Bot Memo classified between January 2025 and July 2026, 58.9% went to companies where artificial intelligence is the product. The other 41.1% went to companies that added AI to something that already worked, sold infrastructure to AI builders, or had no meaningful AI at all.
AI washing is the practice of exaggerating or fabricating AI capabilities to win funding, customers, or a higher valuation. It has become the credibility problem of this funding cycle. The SEC has fined firms over it, and the FTC built an enforcement sweep around it.
Spotting it before the term sheet is a screening skill, and it comes down to a handful of checks a partner can run in an afternoon. What follows is the checklist: seven red flags, ten questions for technical diligence, and the enforcement cases that show what exaggerated claims about AI cost the companies that made them.
On this page
- What Is AI Washing? A Definition for Investors
- 7 Red Flags That Signal AI Washing
- AI Native vs AI Augmented vs AI Adjacent: A Classification Framework
- Due Diligence Questions That Expose AI Washing
- Real Examples: AI Washing in the Wild
- The Regulatory Crackdown on AI Washing
- FAQ: AI Washing Questions Answered
What Is AI Washing? A Definition for Investors
AI washing is the artificial intelligence version of greenwashing. A company overstates, fabricates, or misrepresents its use of AI to look more advanced than it is. The spectrum runs from a startup calling a rules-based workflow “AI-powered” to a public company rebranding legacy software as machine learning without touching the underlying technology.
Sorted by the role AI plays in the product, the 9,214 classified rounds break down like this:
| Classification | Rounds | Share | What It Means |
|---|---|---|---|
| AI Native | 5,427 | 58.9% | AI is foundational. The product would not exist without it |
| AI Augmented | 2,531 | 27.5% | An existing product with AI features added |
| AI Adjacent | 990 | 10.7% | Enables AI (infrastructure, data, tooling) but does not run models in-product |
| AI Platforms | 221 | 2.4% | Trains foundation models (large-scale AI systems like GPT-4 or Gemini) or designs AI chips |
| Non-AI | 45 | 0.5% | No meaningful AI despite the positioning |
Bot Memo classification of 9,214 AI funding rounds, January 2025 to July 2026.
AI washing is not one of these categories. It is a mismatch between them: a company that fits the AI Augmented description while pitching itself as AI Native. Both are legitimate businesses. Only one of them earns the AI Native multiple.
7 Red Flags That Signal AI Washing
Seven signals show up before a single diligence call, and every one of them is checkable from public material.
1. No Technical Specificity
The company describes its technology as “AI-powered” or built on “proprietary algorithms” and never names the approach. Companies with real AI talk about model architectures, training methods, and evaluation. AI buzzwords with no technical nouns behind them are a red flag.
2. AI Bolted On After ChatGPT
Pull the company’s site history. If artificial intelligence went unmentioned before ChatGPT launched in November 2022, and the core product had shipped for years without it, the AI layer is probably cosmetic.
3. No ML Engineers on the Team
Read the team page. A company building AI systems needs machine learning engineers, data scientists, or research scientists. A technical team made up entirely of application engineers points to a wrapper around someone else’s model API.
4. Cannot Explain Training Data
Ask where the training data comes from. Companies with genuine AI describe collection, annotation, and dataset composition without hesitating. Companies engaged in AI washing say “proprietary data” and stop there.
5. The Product Works Fine Without AI
Remove the AI component. If the product still delivers its core value, AI is a feature rather than the foundation. That single test draws the line between AI Native, AI Augmented, and AI Adjacent.
6. “Proprietary Algorithm” With No Public Technical Footprint
Research teams publish papers, file patents, release models, or at minimum talk through their approach at conferences. A claimed AI breakthrough with zero public trace deserves skepticism.
7. AI Absent From Technical Documentation
Read the API docs and developer material. If AI carries the marketing materials and never appears in the technical documentation, ask which one describes the product.
Real example: DoNotPay sold itself as “the world’s first robot lawyer.” The FTC’s 2024 complaint alleged the company did not test whether its chatbot’s output was equal to the level of a human lawyer, and did not hire or retain any attorneys. The order was finalized in February 2025 with $193,000 in monetary relief.
AI Native vs AI Augmented vs AI Adjacent: A Classification Framework
Most diligence frameworks treat AI as binary. A company either “uses AI” or it does not. Bot Memo’s four-part classification system sorts by the role AI plays in the product’s existence, which is the part that moves valuation.
AI Native (58.9% of classified rounds)
AI is the product. Strip it out and nothing remains. These companies carry ML engineers as a meaningful share of the technical team, own training data or a data loop, and treat model performance as a core metric.
Example: Perplexity, an answer engine whose responses are generated by large language models.
AI Augmented (27.5% of classified rounds)
A real product exists on its own and AI improves it. The AI adds value without being load-bearing, and the core product usually predates the AI layer by years.
Example: Canva, a design platform whose editor shipped years before its generative AI features. Turn the AI off and the editor still works.
AI Adjacent (10.7% of classified rounds)
Enables AI without running models in-product. Data infrastructure, labeling, MLOps (machine learning operations) platforms, and chip companies sit here. They sell picks and shovels, and their claims are usually the most literal in the market.
Example: Scale AI, which prepares training data for other companies’ models.
AI Platforms (2.4% of classified rounds)
Trains foundation models or designs the silicon they run on. At 2.4% of classified rounds, it is the smallest group in the market by a wide margin.
Each tier carries a different risk profile, margin structure, and moat. A company priced as AI Native while operating as AI Augmented is a valuation error, not a taste difference, which is why the classification belongs in the screening memo rather than the post-mortem.
Due Diligence Questions That Expose AI Washing
Ten questions separate genuine AI capabilities from marketing hype. Ask them in technical diligence, not in the management presentation.
-
What specific AI models does the product use, and are they trained in-house or licensed? A company making off-the-shelf API calls is AI Augmented at best.
-
Show me model performance over the last 12 months. Real systems track precision, recall, and F1 (a combined measure of accuracy and completeness), or a domain metric. No metrics usually means no models.
-
What share of the engineering team works on ML? Ask for headcount, not a percentage the CEO estimates on the call. A team where ML is a rounding error is not building AI as the product.
-
Describe the training data pipeline. Genuine AI teams walk through collection, cleaning, annotation, and versioning without preparation.
-
What happens to the product if the AI comes out? One question, and it settles whether AI is foundational or decorative.
-
How do you handle model drift and retraining? Models degrade. Teams with real AI have monitoring and a retraining cadence. Teams without it usually have static rules.
-
What is AI infrastructure as a share of COGS (cost of goods sold)? Real inference and training spend show up in the cost line. A negligible number is a tell.
-
Does performance improve as data volume grows? Machine learning systems get better with data. A rule-based system does not.
-
Any published research or filed patents on the AI? Not mandatory, and a total absence across a team claiming a breakthrough still counts as a signal.
-
Walk me through one customer outcome where AI was the difference. Ask for a measurable result. Qualitative answers every time means the AI impact is unproven.
Real example: Delphia claimed in its SEC filings, in a press release, and on its website that it put collective client data to work so its AI could “predict which companies and trends are about to make it big.” The SEC’s order found those statements false and misleading, because the firm did not have the AI and machine learning capabilities it claimed.
Real Examples: AI Washing in the Wild
AI washing is not a thought experiment. Regulators have brought cases, named companies, and collected penalties.
SEC Enforcement: Delphia and Global Predictions (March 2024)
The SEC’s first AI washing enforcement actions hit two investment advisers:
- Delphia (Toronto) claimed from 2019 to 2023 that it used AI and client data to predict market trends. The SEC found the claims false. Penalty: $225,000.
- Global Predictions (San Francisco) marketed itself as “the first regulated AI financial advisor” offering “expert AI-driven forecasts.” The SEC found the claims false and misleading. Penalty: $175,000.
The combined $400,000 is small money next to the precedent. Materially false AI claims by a regulated adviser can violate the antifraud provisions of federal securities law.
SEC Charges Against a Startup Founder (April 2025)
The case closest to a venture portfolio came in April 2025, when the SEC charged Albert Saniger, founder and former CEO of shopping app Nate, over more than $42 million raised on AI claims. The complaint alleges that Saniger told seed and Series A investors the app used automated technology to complete purchases without human involvement, when it relied in large part on contract employees keying in orders by hand. The charges have not been resolved in court.
That is the version of AI washing that reaches a cap table rather than a consumer.
FTC Operation AI Comply (September 2024)
The FTC opened “Operation AI Comply” in September 2024 with five actions against companies using AI hype to sell everything from fake reviews to online storefronts. The sweep carried on: the FTC sued Air AI Technologies in August 2025 over earnings and refund claims made to small business buyers, and in March 2026 the agency announced a proposed settlement that bars the company and its three owners from marketing business opportunities and carries an $18 million judgment, largely suspended on inability to pay.
Facial Recognition Accuracy Claims (January 2025)
The FTC alleged that IntelliVision could not support its claims that its facial recognition software carried zero gender or racial bias and one of the highest accuracy rates on the market. The order finalized in January 2025 bars those claims unless competent and reliable testing backs them.
Each of these claims sounded specific until someone asked for the test results.
The Regulatory Crackdown on AI Washing
AI regulation now has teeth on both sides of the Atlantic. The SEC and the FTC police AI claims in the United States, and the EU AI Act adds a third set of obligations for anyone selling into Europe.
SEC (United States)
The SEC named advisers’ use of AI in its fiscal 2024 examination priorities. The Delphia and Global Predictions orders set the precedent: misrepresenting AI capability in filings, press releases, or on a website violates the antifraud provisions of federal securities law.
For startups, that reaches into fundraising decks and investor updates.
FTC (United States)
The FTC’s reach is broader than the SEC’s. It covers any company making deceptive AI claims to consumers, not only registered advisers. The agency stated its position when it announced the sweep: “there is no AI exemption from the laws on the books.”
The recurring targets are companies claiming capabilities the technology cannot deliver, putting “AI” in a product name with nothing behind it, and tying earnings promises to AI features.
EU AI Act (European Union)
Article 50 of the EU AI Act applies from 2 August 2026. Providers have to design systems so people are told when they are interacting with an AI system directly, and have to add machine-readable marks to AI-generated or manipulated content. Deployers carry their own duty to disclose emotion recognition and biometric categorisation tools, deepfakes, and AI-generated text published on matters of public interest without human review.
For companies selling into Europe, an inflated AI claim stops being only a marketing problem.
FAQ: AI Washing Questions Answered
What is AI washing and how can lawyers prevent it?
AI washing is a company misrepresenting its use of artificial intelligence. Counsel can prevent it by testing every public AI claim against what the technology actually does, substantiating AI statements in filings and investor materials, and requiring technical sign-off before a claim ships. Delphia and Global Predictions agreed to $400,000 in civil penalties over claims that did not survive that test.
As a layperson, how can I recognize AI washing?
Three signals carry most of the weight: the company cannot say which problem AI solves in the product, the AI feature arrived recently on a product that worked without it, and the description never gets more specific than “AI-powered.”
Are many companies overstating their use of AI?
Across 9,214 classified funding rounds from January 2025 to July 2026, 27.5% went to companies that added AI to an existing product and 0.5% went to companies with no meaningful AI behind the positioning. Neither number measures deception on its own. They show how many companies would have to describe themselves carefully to be accurate.
How can you tell if a platform’s AI is real?
Ask for a technical demo instead of a sales demo. Request model performance metrics, training data documentation, and a look at the ML infrastructure. Companies with genuine AI solutions can prove their AI in an hour. Companies without them will offer a case study instead.
Why do companies exaggerate their AI capabilities?
Valuation. An AI label attracts higher multiples, more investor attention, and more coverage. When 58.9% of funded rounds go to companies where AI is the product, the rest face pressure to sound like they belong in that group.
How does AI washing affect consumers and businesses?
It erodes trust in genuine AI solutions, pushes capital toward companies that market well rather than build well, and sets product expectations no rules engine can meet. Once the SEC or FTC gets involved, it becomes a legal and financial liability.
What should I ask vendors about their AI claims?
Open with “what happens to the product if you remove the AI?” Then work through training data, model architecture, ML team size, and one measurable customer outcome. The ten questions above run in a single technical session.
How does AI washing affect innovation?
Capital misallocation is the real damage. Only 2.4% of classified rounds went to companies training foundation models or designing AI chips, the most capital-hungry group in the market.
About the author
Editorial Staff
The Editorial Staff at Bot Memo is a team of writers, analysts, and AI agents dedicated to mapping the global AI startup ecosystem. Led by Chintan Zalani, the team tracks thousands of funding rounds, classifies companies across verticals, and distills it all into actionable intelligence for investors and founders.
