# The Information Asymmetry That AI Created in Its Own Market
## Seminal Perspectives | Touch Stone Publishers | WP Category 546
—
There is a paradox at the center of the current AI acquisition market that most of the participants are too busy to notice. The firms that adopted AI-powered due diligence tools earliest, moved fastest, and processed the most deal flow have eliminated the information advantage they were trying to create. When 86% of PE firms use the same GenAI tools to review the same categories of targets, the speed advantage of AI due diligence has been fully commoditized. The firms are not faster than each other. They are faster together, processing more deals, competing harder, and paying higher premiums for assets that their AI tools have validated — but not actually audited.
This is not the information asymmetry the tools were supposed to produce. It is the opposite.
The original logic was sound. AI tools would allow firms that adopted them to process more information faster, surface better targets earlier, and complete diligence on more deals with fewer people. The firms that moved first would have an analytical edge over the market. What the logic did not account for was that the AI tools are symmetric: the same tools, reviewing the same documents, producing similar outputs for all buyers. The analytical edge disappears when everyone has the same edge. What remains is a collective confidence in a process that has a specific and significant blind spot.
The blind spot is this: AI due diligence tools are pattern-recognition systems applied to documents. They cannot validate whether the AI system described in those documents performs as described. They cannot assess whether the training data that produced the AI’s capabilities was legally acquired. They cannot detect a gap between a company’s AI capability claims and its actual technical architecture. They are analyzing the layer of documentation that sits on top of the technical reality. The technical reality is invisible to them.
The Nate Inc. case made this concrete in a way that no abstract argument could. The DOJ established that a company documented an autonomous neural network capability that its production system did not have. The documentation was coherent. The claims were specific. The VDR was reviewable. Every AI due diligence tool that reviewed the documents found what was in the documents. No tool found what was not there, because no tool was looking at the code.
The information asymmetry that now exists in the AI acquisition market is not between sophisticated AI-using buyers and unsophisticated non-using buyers. That asymmetry has been commoditized away. The asymmetry that remains is between buyers who have added the algorithmic audit to their process — who look at the code, test the AI against its claimed performance benchmarks, assess the training data provenance — and buyers who have not. The former have actual information about what they are acquiring. The latter have document-level confidence about a technical reality they have not examined.
This distinction matters now in a way it did not matter three years ago, for two specific reasons.
The first is regulatory. The EU AI Act assigns parental liability to acquiring entities for the compliance posture of AI systems in their portfolios. R&W insurers are excluding AI-specific risks from coverage. The uninsured regulatory exposure from acquiring a non-compliant AI system — fines up to 3% of the acquiring entity’s global turnover — is a balance-sheet risk that exists whether or not the buyer knew about the compliance gap at acquisition. Ignorance is not a defense under the Act. The acquiring entity is the responsible party from the moment of acquisition. The algorithmic audit is the mechanism that converts an unknown liability into a known quantity that can be priced into the deal or used as a basis to walk away.
The second is evidentiary. The Delaware Chancery Court ruled in April 2026 that AI-generated logs and chatbot records are admissible as evidence of intentional dishonesty in earnout disputes. This ruling changes the litigation environment for post-close disputes in AI acquisitions in a specific way. The AI system’s own records are now discoverable. If an acquired AI system’s performance logs contradict the capability claims that supported the acquisition thesis, those logs are available to any party challenging the earnout protections or the acquisition representations. The buyers who validated the AI’s performance pre-close with an independent algorithmic audit have a factual baseline that they set before the deal closed. The buyers who did not have a factual baseline set by the AI’s own records — records they never saw.
The philosophical question underneath all of this is old, even if the surface problem is new. How does a fiduciary act when the tools designed to reduce information asymmetry have created a new form of it? The answer has not changed across centuries of governance thought: the fiduciary goes one level deeper. When everyone is reading the same documents and trusting the same analytical layer, the fiduciary advantage lies in examining what that layer is sitting on.
In 1844, railway investors were similarly positioned. They had prospectuses, engineers’ reports, and surveys. The information they were missing was whether the railway would actually be built to the standard described. The investors who went to see the construction sites did better than the investors who reviewed the prospectuses. The site visit was not faster. It was not cheaper. It was the level of diligence that the asset required.
The algorithmic audit is the site visit for AI acquisitions in 2026. It is not faster than document review. It is not cheaper. It is the level of diligence that these assets require — and the firms that build it into their standard process will have an information advantage that is genuinely asymmetric, because it is not available from the same tools everyone else is using. It requires different expertise, a different process, and a different understanding of what the AI acquisition risk actually is.
The firms that capture this advantage will not capture it by being first. The window for first-mover advantage in AI acquisition speed has closed. They will capture it by being the ones who looked at the code when everyone else was looking at the documents.
—
*WP Category: Seminal Perspectives (546) | TSP_2026-020 | May 2026*