The number comes from Bain & Company, in its Global M&A Report 2026: one in five strategic acquirers say they walked away from a deal in 2026 because of the anticipated impact of AI on the target. Over the same period, the use of AI in M&A processes more than doubled, reaching 45% of practitioners. AI has shifted from a growth argument to a standalone risk item in due diligence.
The question every executive should be asking: what, precisely, makes a sophisticated buyer back off? The answer isn't "AI is risky." It's more specific, and more mechanical.
This is the principle experienced buyers know by heart, because they've already paid for it. In 2017, after the disclosure of massive data breaches at Yahoo, Verizon cut its acquisition price by 350 million dollars — from 4.83 to 4.48 billion. The risk wasn't in the contract. It was in the asset. And it had to be repriced.
Same logic with Marriott: by acquiring Starwood, the group inherited a breach undetected in due diligence — and an 18.4 million pound GDPR fine. The UK regulator explicitly faulted a failure to verify at the point of acquisition. Regulatory liability doesn't stop at closing. It travels with the asset.
The EU AI Act sets up exactly this context for AI assets. The obligations of Articles 9 to 15 (risk management, technical documentation, human oversight, robustness) attach to the system. The penalties of Article 99 reach up to 35 million euros or 7% of worldwide turnover. A buyer who hasn't priced them in discovers them later — when it's too late to negotiate.
The observation isn't a seller's opinion. It's documented by the players who structure transactions.
a leading audit firm (December 2025) is, to our knowledge, the only leading audit firm to link the AI Act explicitly to value: adhering to "GDPR, the EU AI Act… prevents legal exposure… that could affect the valuation." Brown Rudnick goes further on the exit question: "compliance may become the golden ticket to a successful exit; noncompliance, a market access barrier." On the insurance side, Skadden notes that W&I insurers are now scrutinizing AI issues — to the point of excluding certain representations from the policy. And Reed Smith observes that AI risk is already translating into deal mechanics: dedicated escrows "of 18 to 24 months."
As for the magnitude of the repricing, FE International quantifies it in its 2026 valuation model: regulatory, privacy and technical risks can reduce AI valuation multiples by 15 to 30%. Their most telling case study: a consumer AI target discounted by 25% despite strong growth, for regulatory exposure and the absence of explainability controls.
This is the nuance sellers grasp too late. A sophisticated buyer never simply "overlooks" an unresolved regulatory risk — they turn it into money, in three ways: a direct haircut on the multiple, an escrow that holds back part of the price until compliance is proven, or, beyond a threshold, a walk-away. In all three cases, it's the seller who picks up the tab.
The only variable the seller controls: arriving in the deal room with compliance already established and proven. Established twelve months before the transaction, it costs a fraction. Handled during the deal, under buyer pressure, it costs far more — when it doesn't simply scare off the fourth bidder.
One point of honesty, because it's what separates a defensible argument from an attackable one: none of these sources quantifies a discount attributed by name to the AI Act alone. The 15-30% covers AI risk in the broad sense — regulatory, data, technical — of which the AI Act is an explicit and growing component. So the accurate framing isn't "the AI Act costs you 30%," but: the risk of (non-)compliance with the AI Act has become a recurring factor in the repricing and structuring of AI deals. That's already substantial. And it's verifiable.
For a fund or an acquirer: bring AI Act compliance into due diligence, on the same footing as legal and financial — before the LOI, not after closing. For a target preparing to raise or sell: establish its level and prove it, because it has become a condition of access to the best buyers and the best multiples. In both cases, the gap can be measured, and it can be worked.