Tyler Shultz, a central whistleblower in the Theranos scandal, is warning that AI creates difficult conditions for investors trying to verify a startup’s technology. In an interview published by The San Francisco Standard on October 30, 2026, he connected the AI funding boom with a familiar danger: founders raising substantial sums before investors understand what actually works.
His concern centers on a distinction that a polished demonstration can leave unresolved:
“It’s really hard to tell the difference between something truly unique versus a ChatGPT wrapper,” Shultz told The Standard.
TechCrunch’s coverage that day also reported his warning that large investments give founders tremendous power, making scrutiny consequential well beyond the investment itself.
Shultz’s remarks are an attributed warning about an investment vulnerability, not evidence that any named AI company is fraudulent. The published reporting contains no technical audit of an AI startup. The practical diligence framework below is analysis of his remarks and the documented Theranos precedent.
The Theranos Lesson Is About Access to Evidence
Theranos reached a peak valuation of around $10 billion despite serious failures in its blood-testing technology, according to TechCrunch. Its devices reached Walgreens stores and produced faulty results for patients. The company used its own equipment for only a fraction of its tests, relying on third-party equipment for others.
The investment detail is especially relevant to Shultz’s warning. TechCrunch reports that Theranos withheld scientific information and evidence that its machines worked from potential investors, citing trade-secret concerns. That secrecy deterred many leading Silicon Valley venture investors, although the company still raised money from wealthy individuals and families.
The lesson is more specific than “investors should be careful.” A company’s explanation for withholding evidence can become part of its appeal: the technology is supposedly so valuable that outsiders cannot be allowed to inspect it.
Confidentiality can be legitimate. The practical question is whether it limits public disclosure or prevents meaningful verification altogether. Restricted technical access, confidential document review and agreed evaluation procedures are possible ways to protect sensitive information while still allowing scrutiny.
My view is that this distinction matters more than whether a founder appears convincing. A diligence process should produce evidence that survives without the founder in the room. If every substantive question returns to a promise, the investor still has an unresolved claim.
A Wrapper Can Be Useful Without Being Novel Technology
An AI wrapper is an application layer built around an existing model. It can take a user’s request, add instructions or relevant information, send that material to a model and present the result through its own interface.
That description says little by itself about the product’s commercial value. An application could provide useful integrations, specialized data access, workflow controls or a convenient way to complete a difficult task. None of those benefits requires its developer to have trained a new foundation model.
Shultz’s distinction becomes important when the company’s pitch and its actual contribution diverge.
Consider two hypothetical startups showing an identical document-analysis demonstration:
- One claims model-level invention. It says its own training process produces capabilities competitors cannot easily reproduce.
- The other claims application-level value. It uses an outside model but promises better document handling, customer integrations and workflow reliability.
The same successful answer would support a narrow observation: both demonstrations handled that particular example. It would leave several investment questions unanswered, including who supplied the underlying capability and how difficult the product would be to copy.
I'd start diligence with the claimed source of value. A company selling workflow software should be evaluated on the strength of that workflow. A company seeking a premium for proprietary model research needs evidence supporting the research claim.
This also separates two risks that can otherwise get blurred. An honest company can be poorly differentiated and overpriced. A deceptive company can misrepresent what it owns or what its system can do. Investors need enough visibility to distinguish both situations from a defensible business.
What a Demo Shows, and What It Leaves Open
A working demonstration is evidence of something. It shows that a particular setup produced a particular result under the conditions presented.
The mistake is extending that observation too far.
A demo alone may leave open whether the system can handle unfamiliar inputs, how often it fails, how much human assistance it needs and whether the economics work at customer scale. A smooth interface also cannot, by itself, establish whether the underlying capability comes from proprietary technology or an external service.
These are separate questions:
| Investment claim | Evidence worth requesting | What it helps clarify |
|---|---|---|
| The company owns important model technology | A confidential architecture review and a clear account of external dependencies | Which parts of the system the company actually controls |
| The product outperforms alternatives | Comparisons on the same tasks, with documented settings and evaluation criteria | Whether the advantage survives a fair comparison |
| The system generalizes beyond the demo | Tests on withheld examples and unfamiliar customer inputs | How performance changes outside selected cases |
| The product scales economically | Measured usage costs, human-review requirements and customer-level margins | Whether successful outputs can become a sustainable business |
| The data creates a defensible advantage | Documentation of data sources, rights, access and permitted uses | Whether the claimed advantage is usable and durable |
This is a proposed diligence framework, not a checklist supplied by Shultz. Its logic follows from his warning: if the visible output cannot reveal where the value comes from, investors need evidence from other parts of the system.
Technical Diligence Should Match the Claim
A startup claiming to train its own models invites questions about training-data provenance, model ownership and the evaluation process. An application company relying on an outside model calls for a different examination: supplier dependencies, integration quality, data handling and the contribution made by its own software.
Requesting every possible technical artifact from every startup would waste time. The useful approach is to connect each important investment claim to something observable.
Frequently Asked Questions
4 questions
1Who Is Tyler Shultz?
Tyler Shultz worked at Theranos and became a central whistleblower who helped expose problems with the company’s blood-testing technology.
