Surveillance Pricing Meets AI: Why Regulators Are Suddenly Scrutinizing Data-Driven Price Personalization and What It Means for Privacy

If you ship pricing models that touch personal data, the regulatory ground shifted under you in 2026. What regulators now call "surveillance pricing" (using browsing history, location, demographics, or behavioral inferences to set individualized prices) has moved from a niche consumer-advocacy complaint to an active enforcement target at the federal and state level. The core issue is privacy: not just whether you collect the data, but whether your architecture can prove that a given price was derived from aggregate market signals rather than from an individual's inferred willingness to pay. That distinction, which sounds like a policy subtlety, is actually an engineering problem. And most pricing stacks were not built to make it legible.

Key Takeaways

• The FTC, at least three state legislatures, the California AG, and the House Oversight Committee have all moved against AI-driven individualized pricing since mid-2024, with Maryland enacting the first outright ban in April 2026.
• The regulatory line being drawn is between "dynamic pricing" (responding to inventory, demand, time-of-day) and "surveillance pricing" (responding to who the buyer is). If your model ingests customer-level features, you are on the surveillance side of that line until you can prove otherwise.
• The real liability surface is often not your own model but your third-party data supply chain: the intermediary vendors whose segmentation signals feed your pricing algorithms are exactly what regulators are subpoenaing.
• Consumer trust data now backs the compliance case commercially: surveys show 71% of consumers say personalized pricing makes them prefer in-store shopping, and 44% would avoid a brand whose pricing lacked transparency.
• With 40+ bills across 24+ states, ad hoc compliance is already unsustainable. Privacy-preserving pricing architecture (aggregate-only signals, provable non-discrimination) is becoming a structural requirement, not a nice-to-have.

What Is Surveillance Pricing, Exactly?

Surveillance pricing is the practice of using detailed personal data to set individualized prices for the same product or service. The FTC's July 2024 orders targeted eight companies whose products used "advanced algorithms, AI and personal data" to set prices tailored to individual consumers. The data inputs cited: location, demographics, credit history, browsing behavior, shopping history. The key word is "individual." Dynamic pricing that responds to aggregate supply and demand (surge pricing on a ride-share app when demand spikes citywide, or a hotel raising rates when occupancy crosses a threshold) is a different animal. Surveillance pricing responds to you, specifically: your zip code, your device, your past purchase cadence, your inferred price sensitivity.

Machine learning made this scalable in a way that manual price discrimination never was. A January 2025 FTC staff research summary, based on documents obtained through a 6(b) inquiry, confirmed that precise location and browser history are frequently used to give individual consumers different prices for identical goods. The problem, from a regulatory standpoint, is that consumers cannot easily detect it. You see your price. You do not see everyone else's.

Why Did Regulators Move So Fast?

Because the evidence went from theoretical to concrete. In December 2025, Consumer Reports (working with the Groundwork Collaborative) found that Instacart was charging some shoppers up to 23% more for identical items from the same store at the same time. That is not a rounding error. That is a measurable, reproducible price gap tied to the identity of the buyer. The FTC, which had already secured a separate $60 million judgment against Instacart for unrelated deceptive practices, turned its attention to the company's AI-driven pricing mechanisms.

Then the legislative branch got involved. On March 5, 2026, the House Oversight Committee launched a formal investigation, sending letters to major travel and platform companies requesting documentation on revenue management algorithms, data inputs to pricing, and A/B testing practices. Separately, on January 27, 2026, California AG Rob Bonta announced an investigative sweep into businesses using personal information to set targeted, individualized prices, covering retail, grocery, and hotel sectors.

The velocity is notable. Two years ago, there were zero enacted surveillance pricing laws. Now there are three states with enacted or passed legislation, a fourth with a bill through one chamber, and 40+ bills across at least 24 states.

What Laws Have Actually Been Enacted?

Maryland moved first. On April 28, 2026, Governor Wes Moore signed the Protection from Predatory Pricing Act, making Maryland the first state to ban the use of surveillance data to raise prices in certain industries. Violations carry fines up to $10,000 per incident, or $25,000 for repeat violators. On June 4, 2026, New York's legislature passed the One Fair Price Act (pending the governor's signature), and on the same day Connecticut's governor signed a law prohibiting retailers and delivery services from engaging in surveillance pricing.

California's AB 2564 passed the California Assembly on May 27, 2026, and is pending in the state Senate. Illinois and Vermont are pursuing their own versions. The approaches vary: some states require disclosure, others impose outright prohibition. If you operate nationally, you are looking at a patchwork that will only get denser.

At the federal level, two bills are pending (the One Fair Price Act and the Stop AI Price Gouging and Wage Fixing Act), but neither appears close to passage. The investigative pressure from the FTC and House Oversight is real; the statutory authority is still catching up.

Is the FTC Itself Aligned on This?

No. The agency's own commissioners have publicly disagreed. A dissenting statement from then-Chairman Andrew Ferguson, joined by Commissioner Melissa Holyoak, pushed back on the Surveillance Pricing 6(b) staff research summaries, questioning the framing and conclusions. This matters for you as an operator because it signals that the regulatory posture could shift with future commission composition. But the state-level momentum is largely independent of FTC internal politics. Even if the FTC softens its approach, you still face the California AG, the Maryland statute, and 20+ other state legislatures that are not waiting for federal consensus.

How Should You Think About the "Dynamic vs. Surveillance" Distinction?

This is the single most important compliance question if you ship pricing models. Regulators are drawing the line clearly: dynamic pricing that responds to market-level conditions (inventory levels, aggregate demand curves, time-based scarcity) is generally lawful. Surveillance pricing that responds to individual consumer characteristics (inferred income, browsing history, purchase frequency, location as a proxy for willingness to pay) is the target.

The problem is that most pricing systems blur this line at the architecture level. Your demand model might ingest both aggregate signals (total inventory remaining, regional demand index) and individual signals (this user's session count, their device type, their referral source) through the same feature pipeline. From the model's perspective, these are all just features. From a regulator's perspective, they are categorically different, and the burden of proving which features drove a given price is on you.

This is not a PR problem you solve with a transparency page. It is an architecture problem. Can your system produce, on demand, an audit trail showing that a specific price was set from aggregate or market-level inputs, without individual consumer characteristics influencing the output? If the answer is "not without significant refactoring," that is the work.

What Does "Provable Separation" Look Like in Practice?

At minimum, it means your pricing pipeline has a clear, auditable boundary between market-level signals and individual-level signals. A few concrete patterns:

• Pricing models that accept only cohort-level or aggregate features as inputs, with individual identifiers stripped before inference.
• Logging infrastructure that records, for every price served, the exact feature vector that produced it, so you can demonstrate post hoc that no individual-level data was present.
• Separate model registries for "market-responsive" and "customer-responsive" pricing, with the latter subject to explicit legal review before deployment in any jurisdiction.

None of this is trivial. But the alternative, a monolithic pricing model that mixes aggregate and individual signals and cannot explain which drove any given output, is precisely the system that regulators are targeting.

Where Does the Real Liability Live?

For many operators, the highest-risk surface is not your own model. It is your third-party data supply chain. The FTC's original 2024 orders targeted intermediary companies that sell "pricing-as-a-service" built on purchased third-party data. These vendors aggregate consumer data from data brokers, infer price sensitivity or willingness to pay, and pipe those signals into your pricing engine. From your perspective, you are calling an API. From a regulator's perspective, you are using surveillance data to set prices.

If you buy pricing intelligence from a vendor, the question you need answered is: what data sources feed the segmentation signals I receive? If the answer involves purchased browsing history, inferred demographics, or location data that was not collected with explicit consumer consent for pricing purposes, that is your exposure. The House Oversight letters specifically requested documentation on data use in pricing and testing practices. That includes vendor data. Your contract with a pricing vendor does not insulate you from the regulatory obligation to know where your inputs come from.

Do Consumers Actually Care?

Yes, and the data is specific enough to be commercially relevant. A Quad/Harris Poll survey from April 2026 found that 71% of respondents said personalized online pricing makes them want to shop in stores instead, where everyone pays the same price. A separate Zamplia survey from February 2026 found that 71% of consumers have noticed AI being used while shopping, and 44% would feel uncomfortable or avoid a brand if pricing lacked transparency.

These are not abstract sentiment numbers. They translate to channel shift and brand avoidance. If your personalized pricing model increases average transaction value by 3% but drives 10% of price-conscious customers to a competitor or a physical store, you may be net negative. Most teams running these models do not measure the counterfactual well enough to know.

What Is Happening in Europe?

The EU is moving along a parallel track. The draft Digital Fairness Act, expected mid-2026, follows a Commission "Fitness Check" that found personalized pricing can be problematic, particularly when firms lack transparency or exploit consumer vulnerabilities. The UK government has separately signaled it may give the Competition and Markets Authority stronger powers to investigate pricing algorithms. If you operate across borders, the compliance surface is not just 24 US states. It is a transatlantic patchwork with different definitions, different enforcement mechanisms, and different burdens of proof.

The European Journal of Law and Technology has published analysis framing algorithmic pricing as a data protection issue under GDPR, not just a competition issue. That matters because it means pricing decisions based on personal data could trigger data subject access requests, legitimate interest challenges, and purpose limitation arguments that are entirely separate from any pricing-specific legislation.

What Should You Do If You Ship Pricing Models Today?

Start with an honest inventory. Map every data source that feeds your pricing pipeline, including vendor-supplied signals, and classify each as "market-level" or "individual-level." If you cannot make that classification for a given input, that is your first audit item, not your last.

Then ask whether your system can produce, for any given price, a record of which features generated it. If you are running a black-box model that mixes cohort and individual signals without logging the feature vector per inference, you have no way to respond to a regulator's request for documentation. The litigation risk here is real: plaintiff-side attorneys are already framing surveillance pricing as the next frontier of privacy litigation, and discovery in those cases will focus on exactly this kind of technical documentation.

Practically, consider these steps:

1. Audit your vendor contracts. Do your pricing intelligence vendors disclose the provenance of their data? Can they certify that their signals are derived from aggregate market data rather than individual consumer profiles? If not, you are inheriting their regulatory exposure.
2. Separate your feature pipelines. Build an architectural boundary between market-responsive features and customer-responsive features. This does not mean you cannot use customer data for other purposes (recommendations, inventory allocation). It means your pricing model should have a clean, auditable input surface.
3. Log everything. For every price served, record the feature vector, the model version, and the timestamp. This is your audit trail. Without it, you cannot demonstrate compliance in any jurisdiction.
4. Monitor the state-level tracker. The landscape is moving fast. A bill that was in committee last month may be on the governor's desk next month. If you operate in multiple states, you need someone watching this weekly, not quarterly.
5. Stress-test your disclosure posture. Several state proposals require disclosure when personalized pricing is used. Even if your jurisdiction does not mandate it yet, consider whether you can explain your pricing methodology to a consumer who asks. If the honest answer is "we cannot," that is a problem independent of any specific statute.

Is Privacy-Preserving Pricing Architecture a Competitive Advantage?

We think so, but we will be specific about why. The argument is not that privacy compliance is inherently profitable. It is that the cost of ad hoc compliance across a patchwork of 24+ state laws, each with slightly different definitions and requirements, is high and rising. Building a pricing architecture that is structurally compliant (aggregate-only signals, provable non-discrimination, auditable feature pipelines) amortizes that cost across all jurisdictions. You pay the architectural cost once. The alternative is paying the compliance cost repeatedly, in every state, every time a new bill passes.

There is also the consumer trust dimension. When 44% of consumers say they would avoid a brand over opaque AI pricing, the ability to say "our prices respond to market conditions, not to your personal data" is a differentiator. Not because it is a marketing message, but because it is a verifiable architectural claim. The brands that can make that claim credibly will have an advantage over those that cannot, particularly as consumer awareness of surveillance pricing grows.

What We Build, and What We Don't Claim

We build Selina, a privacy-first AI assistant. We are not a pricing platform. But the principles underlying this regulatory shift are ones we think about constantly. Selina's memory is encrypted at rest. It is NOT end-to-end encrypted, because a slice of each request reaches a frontier provider at inference. We state that plainly because honesty about architectural limits is how trust actually works. Files and transfers via SelinaSEND are zero-knowledge encrypted. Operational metadata is kept for a short retention window. Your account is protected, your content is encrypted.

We use a stack of frontier models, routed per task. We do not sell your data. We do not use it for pricing. We do not use it for anything except serving you.

The reason we mention this in a piece about surveillance pricing is that the underlying principle is the same: if your architecture cannot prove what it does and does not do with personal data, your policy language is irrelevant. Regulators are not reading your privacy policy. They are subpoenaing your data pipelines.

Where Does This Go Next?

The trajectory is clear, even if the specifics are uncertain. More states will pass surveillance pricing laws. The definitions will vary. Some will require disclosure; others will ban the practice outright. The EU will add its own requirements through the Digital Fairness Act and existing GDPR enforcement. Federal legislation in the US may or may not pass, but federal investigative pressure (FTC 6(b) orders, House Oversight subpoenas) will continue regardless.

If you are building or buying AI pricing tools today, the question is not whether you will need to demonstrate that your prices are derived from market signals rather than individual surveillance data. The question is whether your architecture can do that when someone asks. Because someone will ask. And "someone" increasingly means a state attorney general with subpoena power, not a curious journalist.

Build the audit trail now. Separate your feature pipelines now. Know where your vendor data comes from now. The regulatory window for getting this right before enforcement begins is closing. In Maryland, it already closed.

If you are looking for an AI assistant that takes data architecture and privacy seriously, start a free 7-day trial, no card required.

Michael C.

Founder & Principal Engineer, Selina Labs

Michael builds Selina, a privacy-first AI that remembers you across conversations. He ships security-sensitive AI in production — real attacks, real fixes, measured in minutes and dollars — and writes about privacy, security, and LLMs from that seat. Top Rated Plus and expert-verified on Upwork.

• www.upwork.com
• x.com
• medium.com
• selinalabs.io