Vermont Legislature SmartTranscripts:

[Michael Marcotte (Chair)]: Good afternoon, everyone. This is the Brumohana Committee on Commerce and Economic Development. It is Wednesday, 04/29/2026 at 02:52 in the afternoon. We'll get back from the floor and to continue to take more testimony on S-seventy one, which is the bankruptcy bill. So, I believe we have Ryan who's on first. Good afternoon, Ryan.

[Ryan Krueger]: Okay. Good afternoon. Okay. Let me get started.

[Michael Marcotte (Chair)]: What fancy office you have.

[Ryan Krueger]: Yeah. Well, we'll get there. One second here. Okay. My name is Ryan Krueger. I am a resident of Montpelier, Vermont. I have over 20 experience as an attorney and specialize in consumer protection, privacy, and data security. I have worked for the Vermont attorney general in the FTC's division of privacy and identity protection, and I'm currently the deputy chief of the Massachusetts attorney general's office privacy and responsible technology division. I also teach privacy consumer protection and other classes at the University of Vermont. I testify today in my personal capacity as a citizen of Vermont. My testimony does not reflect the positions of the Massachusetts AG or that of our office. I'd like to begin by thanking the committee for its flexibility and apologizing for the unusual circumstances of my testimony, which are due to a minor heart kerfuffle I experienced yesterday. But I believe that the importance of assisting this committee in any way I can to enact the most protective privacy legislation we can outweighs any minor health concerns. I thought today I'd start with some first principles. Consumer protection law, of which privacy is a subset, just historical accident, is heavily reliant on the paradigm of notice and consent. Basically, consumers should be allowed to make their own informed decisions. Since privacy and data security became a concern, we've tried to address it with a notice and consent regime, and it has been a complete and utter failure. Academics, attorneys general, and the FTC have stated this. And it's not because consumers don't read privacy notices. There are literally not enough hours in the day for consumers to read all the privacy notices. Well known Carnegie Mellon study found that if consumers were to read every privacy policy notice they encountered in a single year, it would take seventy six eight hour workdays to read them all. And even if they could, the notices themselves are usually useless, which makes sense. If the only limitation on a business's collection and use of their data is that it has to line up with what they've written down, then they are going to write broadly worded flexible documents that don't foreclose future opportunities. That's just good lawyering. And even if the notices are clear and specific and brief, they still wouldn't be entirely useful because technology is evolving so fast. Five years ago, no one disclosed that data might be used to train AI models. Now it's common. And finally, the data advertising broker ecosystem is so convoluted that I don't have a complete handle on it. I've been spending my entire career trying to study it. My point is to expect consumers to, you know, derive useful information from notices doesn't make sense. So principle one, notice and consent doesn't work in data privacy. Principle two, we tend to leave things as is. Okay? Were that not the case, four zero one ks plans wouldn't matter if it was opt in or automatic enrollment. They'd have the same enrollment numbers. But we know that automatic enrollment results in like 95% enrollment, and if you put in an opt in regime, it's far lower. Businesses know this. They know that if the response requires an opt in, they will have far less uptake than if it's an opt out. So anytime you put into the law, consumers can opt out of x or y, you're setting up a situation where most people won't do so. Sometimes opt out is unavoidable, but it is not particularly protective. So principle two is requiring consumers to take affirmative acts to protect themselves means that many consumers won't protect themselves because they're too busy, they don't understand the choices, or status quo bias. Now a pure libertarian would say, good. Let them face the consequences of their actions. And in that world, businesses would not have to opt in to having fire insurance, and manufacturers would be allowed to put lead in paint as long as there was a disclosure. But that's not Vermont. Data privacy law is incredibly complicated, but from these two principles we can say that the policy to protect consumers won't rely on notice and consent and won't put the burden on consumers to protect themselves from the extremely sophisticated companies that make billions of dollars from surveilling them and then using that surveillance to manipulate them into making the choices that are against their own interest. Another way to put this is a good law will put obligations on businesses to adhere to a regulatory framework without making consumers have to ask them to. It makes the default privacy and puts the onus on the company that wishes to invade that privacy. Now, if there's one silver bullet to solve the problems of privacy and, by the way, data security, it's data minimization. It's a very simple concept. Only collect as much information as you need and no more. After you no longer need the data, destroy it. So if I order a book online from a bookstore, the company needs my payment information, my shipping address, and the name of the book I want to buy. That's it. They don't need to know what kind of computer I'm using, my geolocation information, all the other books I looked at while I was on their site, my job, anything else. Now there are three key advantages to data minimization. One, it's the most privacy protective regime we have because it just limits the amount of data about us that is available out in the ecosystem. Two, it puts the entire burden of compliance on the business and asks nothing of the consumer. The consumer need not take any affirmative action to ward off undue surveillance. And then three, as an added bonus, let's talk for a second about data breaches. It is well known that it is impossible to guarantee that our data is completely safe. Companies will continue to have data breaches. Bad guys will continue to steal our data. If China or North Korea or Russia wants to get into your systems, they will get into your systems. Companies can invest more or less in data security, but they cannot absolutely protect that information. It is not in their control. The amount of data that a company collects and retains, however, is completely within the company's control. So if you can't stop breaches entirely, you can at least minimize the harms that will arise from them. Now data minimization is not a regulatory invention. It was considered a best practice in the business community before legislators started trying to enforce it. Then we have a second concept that goes with data limit data minimization, which is use limitation, which just means that if you collect data for a specific purpose, you may not use it for a different purpose. If you collect someone's phone number for providing two factor authentication, you may not use that number for telemarketing. This concept is critical because we are moving into a world where people are being forced to give more and more data to companies for authentication, for age verification, and for other reasons. And one of the biggest worries now is how the companies will misuse that data. If this law did nothing but impose strong, enforceable data minimization and use limitation restrictions and nothing else at all, it would provide Vermonters with an exponentially higher degree of protection than they have today. There are, I believe, currently 20 state laws that claim to require data minimization. They don't. They have a section called data minimization, but it is not data minimization. Only one of the laws actually requires data minimization. And that's because all those other laws limit, basically say minimization means limit your data collection to what is necessarily related to a disclosed purpose. But you can disclose any purpose. So then you can collect data related to it as long as you disclose it. And of course, notice and consent doesn't work. We're back to notice consent in that situation. The state of Maryland is the only state with a data minimization requirement that is not tied to a disclosed purpose, but it is tied to what is necessary to provide the product or service that the consumer wants. In other words, they can only collect what they actually need and no more. It would be even better if it was limited to both collection and processing, but regardless, I I I cannot emphasize this enough. If you want to protect Vermonters, adopt the Maryland standard, which is the only standard that is actually data minimization. Now as an aside, California's statute does not have a true data minimization because it's tied to disclosed purposes. But then in the regulations, California clarified that those disclosed purposes have to be consistent with what a consumer would expect when buying the product or service. So in other words, California accomplished in two steps with a lot more verbiage what Maryland accomplished in one. So if for some reason a Maryland law is unacceptable, though I can't fathom a good faith reason why, then you should adopt California's approach. But Maryland was the first state to successfully buck Big Tex push to keep doing whatever they want, and we should be following their lead. Now, I'd love to talk to you about why public data exemptions seriously undermine data privacy laws and are actually a huge loophole, or why enforcement is so critical, or try to convince you why a private right of action is not in fact going to bring about the fall of the republic, or heck, data use level exemptions, but I believe I have used my time. Thank you.

[Michael Marcotte (Chair)]: Thanks, Ryan. Questions for Ryan?

[Ryan Krueger]: Say again?

[Herb Olson (Member)]: So, you weren't gonna talk about enforcement, I'll ask you about it anyway. And it seems to me that you can add the best standards of the world, but if they're not enforceable, they don't mean a whole lot. And it seems to me from listening to your testimony, there's kind of three ways you can go about enforcing something in this sphere. One, as you mentioned, a private right of action. The other is saying to the EAG, go to it guys. There's a third option which would and the issue with sort of directing the AG to do their normal enforcement with the resources that they have is that they don't always have the resources to do that. And some states, some regulatory regimes give that enforcement agency the ability to examine, analyze hire folks, pre litigations in order to review what's going on if they have an allegation of a violation, and then proceed from there. Any thoughts on those three

[Ryan Krueger]: I just want to clarify. I heard private right of action. I heard AG. What is the third? Is it just AG with CID authority? Is that kind of?

[Herb Olson (Member)]: No, because you're right. I mean, that's the typical regime under a consumer protection statute. In other regulatory regimes, insurance, banking, that kind of thing, the regulator has the ability when they are interested in examining a company for whatever reason, they have the ability to not only use their own staff, but to hire other staff for that specific purpose and review, analyze. Sometimes I think I would imagine this stuff takes a little while and some real high expertise to be able to do the job. So that's the third option on posing.

[Ryan Krueger]: Gotcha. Okay. I'll throw in a fourth, and that's criminal penalties. Okay? We won't go there today, but believe it or not, there are actually privacy laws that have criminal penalties. So let me talk about the private right of action. And by the way, you pretty much made all of my arguments for me. So this is kind of easy. But first off, I don't like using the term private right of action or PRA because it just kind of, it kind of abstracts what is actually just common sense. If someone violates a right that is afforded to you in the law, you should be able to get a remedy. You should be able to protect that right. Okay? As I said, you know, last week about a different bill. Imagine a world where if someone hit you with their car and damaged your car, your only remedy would be to go to the attorney general and ask them to get you relief. And if the attorney general is too busy or just, you know, too overwhelmed, then you get no relief whatsoever. Right? Because because there's no private right of action for car crashes. Now if if in that world, you guys tried to create the private right of action, you'd be hearing from the auto insurance industry telling you, oh, that's gonna bring all this litigation. You shouldn't do it because everyone's gonna be litigating. Well, yeah, that's how you enforce your rights, you litigate. Actually, the way you enforce your rights is you ask people to make it right. And if the people on the other side think there's a chance they're gonna be sued or think that, you know, there's actually a legal way of the enforcement, they usually do the right thing. It avoids litigation. It's only the really egregious situations where anyone ever has to go to court. That's just that's just reality. That's just the way it works. Somehow, industry has done an incredible job of demonizing private rights of action and demonizing class actions. Okay? But the fact of the matter is if you want a law that companies take seriously, you need as much enforcement as possible. In a way, class action attorneys class action attorneys often refer to themselves as private attorneys general. Okay? Do they often make a lot of money doing what they do? Yeah. They often make a lot of money doing what you do. You know, here's a secret. The defense lawyers working for the other companies, they often make more money than the plaintiff's lawyers do. Lawyers make a lot of money for doing what they're doing. That's what they they do. Is there frivolous litigation? I mean, sometimes you sue someone and the court says this isn't a valid case. Does that mean it was frivolous or not? I mean, that's just how how lawsuits work. As a general matter, class action lawyers are looking for cases that they can win, not cases that are to get thrown out. I think that the whole shakedown litigation thing is a, you know, a little bit of a boogeyman. Before someone calls out about BIPA, the Illinois Biometric Information Protection Act, I wanna distinguish that law, has a private right of action and $1,000 in exemplary damages or 5,000 for a intentional or reckless violation, I think that if you have the private right of action plus statutory damages, that's a very strong law. That's very enforceable. I might not have a problem with it, but that but what we're proposing does not have any of those. It doesn't have the the statutory damages. It would be just damages, which in a privacy case are not always easy to prove, which is gonna limit where that that situation is possible. Yeah. I think I've I've spoken on, for quite a bit about that. But

[Herb Olson (Member)]: Examination authority.

[Ryan Krueger]: Examination authority. Well, if a law is enforced under, the Consumer Protection Act, the Consumer Protection Act gives the AG CID authority. So that's that. As far as the insurance regime, you know, honestly, it's not my specialty area. I think if you want to give I mean, even AGs sometimes hire outside counsel, but you're talking more hiring outsiders to do the investigation, which, you know, I wouldn't have a problem with other than there'd have to be resources allocated to do it. So that, you know, that that would be that. And then I'll tell you, in terms of in terms of criminal penalties, you know, there there's room for it. Like, you know, if if someone is collecting data for authentication purposes and then misuses that data intentionally, there has to be an intent element in criminal, you know, make it a felony. Because here's the thing, people are freaking out about all these authorization laws and age verification laws because they know that they're gonna have to give more data to companies and those companies will misuse them. And those people are probably right. We are in a regime where we've basically let lawlessness, you know, run rampant. So if people believe that if something goes wrong, there will be real consequences. Someone might go to jail, and that's not crazy. Antitrust has criminal consequences, and we have a lot less price fixing because of it. But that's not in the bill or any of the amendments, so we won't

[Michael Marcotte (Chair)]: worry about that. Questions for Ryan? Okay. Thank you, Ryan.

[Ryan Krueger]: Thank you very much. Have a good day.

[Unidentified Committee Member]: Thanks, Ryan.

[Ryan Krueger]: Thank you.

[Michael Marcotte (Chair)]: Carry out. Good afternoon.

[Ariel Garcia]: Good afternoon. Chair Marcotte and members of the committee, thank you for the opportunity to testify today in support of the Vermont Data Privacy and Online Surveillance Act. My name is Ariel Garcia. I'm chief operating officer of Check My Ads Institute. We are an independent nonprofit digital advertising watchdog, and we advocate for a transparent and fair digital advertising market for advertisers, publishers, and the public. Before joining Check My Ads in 2024, I spent a decade in digital advertising. I most recently served as chief privacy officer at Worldwide, which is a major global advertising agency. For years, I helped advertisers navigate the evolving regulatory environment as they look to strike the balance between privacy, trust, and effective advertising. In fact, I resigned from my role precisely because of my realization that the only winners in the current state of the industry are big tech giants like Google and the data brokers and advertising technology middlemen that thrive in their shadows. The $750,000,000,000 digital ad industry is the main business model of the Internet, and today that business model is broken. Big tech firms like Google and Meta have built empires on unbridled extraction. They've amassed massive troves of data, cemented themselves on all corners of industry and society, unilaterally set norms that benefit their own business to the detriment of their users, advertisers, and publishers alike. And yet while their household names, there are hundreds of other companies in the ad tech sector that have followed in their footsteps. Now before the advent of modern digital advertising, advertisers would typically work with an ad agency to figure out what publications their target audiences are likely to consume. Today, most advertising online is transacted through an automated real time auction, and that's known as programmatic advertising. Over the past two decades, the central promise of programmatic was that it would help advertisers of all sizes more effectively and efficiently reach their audiences, and it would help publishers monetize or sell more of their ad space. But for this to work, advertisers today have to rely on a supply chain of various intermediaries, and that includes ad tech companies, the ad agencies that plan and buy campaigns on their behalf, and data brokers. So for every dollar a business spends on digital ads, each of those middlemen gets a percentage revenue share and most are paid on volume. That creates backwards incentives because every company in the middle stands to gain as more money passes through their pipes. But it's not just money they're getting. They're also getting access to personal data. And that's how programmatic advertising has resulted in constant tracking and widespread sale and leakage of consumer personal data. Now while that process sounds efficient and powerful in theory, the reality is much different. Out of every dollar spent on programmatic advertising, less than half makes it to the publisher. 30¢ is absorbed by those ad tech middlemen. Over 20¢ is wasted on ad fraud, invalid traffic, like bot clicks, and low quality made for advertising websites, think AI spam sites. The average programmatic campaign saw a single advertiser's ads placed on a staggering 23,000 websites. Now those figures, were produced by a leading advertiser trade association, the Association of National Advertisers, and it represents a conservative estimate. It doesn't reflect agency fees or often ineffective fraud prevention tech. It doesn't include spend wasted on advertising that while it technically doesn't meet industry definitions for fraud, doesn't actually adhere to advertisers' criteria. And these statistics are from studies looking at some of the largest brands in the world with the most resources and the most leverage. So while research suggests that ad fraud accounts for a whopping $84,000,000,000 in global ad spend a year, which places it second only to the drug trade as a source of revenue for organized crime. That's just the tip of the iceberg as it comes to advertiser waste, especially for smaller businesses. The reality is that for small businesses, this waste isn't only detrimental to their growth, but at times to their survival. And so we at Check My Ads, in the course of our research, we launched our own experimental ad campaign. We emulated the experience of an average small business user. We set up a new Google Ads account as thousands of other small businesses do. We were nudged into Google's AI driven opaque product, Performance Max. That product is notorious for restricting advertiser transparency and control, has a history of placing ads on potentially illegal content, and also ignoring advertiser targeting criteria. Our findings echoed that research. We found our ads on AI generated spam websites and parked domains. We ran on sites you can't navigate to directly that may have been fraud. We selected to target English language placements. That was one of the only inputs you can configure in these types of campaigns, and we still ended up on foreign language videos. With some of our ads ending up on kids videos, we actually have no idea who Google's algorithm targeted. We ran on videos that are no longer available that were removed for policy violations, but we don't know what policy was violated. Was it adult content? Was it piracy? We simply don't know. There is one place we didn't run. We did not run on one single recognizable publisher website. So obviously, the next question would be, how did it perform? According to Google, it did great. For our modest $75 campaign, we got 34 new sign ups to our newsletter, except we didn't. In our customer database, we only had five new sign ups. So while Google reported this really efficient cost per lead of $2.17, by our math, it was $14.75 per sign up, making our ad spend 580% less efficient than what Google told us. We asked Google about the discrepancy. They told us that they don't have a team dedicated to supporting small nonprofits. So that's one of the companies fighting tooth and nail against strong privacy bills like this one under the pretense that it would hurt Vermont small businesses. Now that's one measly campaign, but this is the default campaign type that the 80% of advertisers using Google Ads are are nudged into. So this is a lot of money that's wasted, especially for small businesses with small budgets. And it's not just Google. Recent reporting revealed that Meta earns 10 of its revenue, so $16,000,000,000 a year from scam ads. And instead of cracking down on the scammers that hurt its customers, it applies an effect of potential scammer tax, and it benefits further from the artificial inflation of ad prices while its users and small business advertisers bear that greater cost. Meanwhile, that meta is nearing a decade of ongoing litigation over inflating advertising performance metrics to the detriment of advertisers. And of course, the same phenomenon occurs outside of big tech's empire. The industry loves to push narratives about the critical need for consumer data for effective advertising. But the reality is that the consumer data that's peddled by data brokers for ad targeting, while it is invasively collected, it's also shockingly inaccurate. Research has shown that leading ad data brokers were right about people's gender only 42% of the time, worse than guessing at random. That is what advertisers are paying for. I've done my own personal desktop research that echoed those findings. One ad platform had me in 500 audience segments, across seven different data brokers. The audiences that I was in were highly inaccurate, often contradictory, and sometimes sensitive. So according to that data, I'm both high income and below the poverty level. I'm listed as a middle aged suburban homeowner without kids, as well as a mother of two. I am neither. I was identified as likely suffering from heart disease, speaking Spanish, and celebrating Ramadan. None of these things are true. While the industry has finally started to acknowledge these data quality realities out loud, that insight is far less accessible to small businesses. Yet right or wrong, that data can be used within and beyond advertising in ways that cause real harm, as you've heard from others. The chief privacy officer of Acxiom, one of the world's largest data brokers, has admitted that their consumer data used by advertisers around the world is comprised of little more than informed guesses. He said he hopes that if they guess wrong, it doesn't have any negative consequences for denial of benefits or credit. But credit reference agencies have purchased and used Acxiom's data to determine creditworthiness. All of this so that data brokers and ad tech companies can sell data that has less than a 50% chance of being accurate to unsuspecting small business advertisers, this is the alleged pinnacle of efficiency that props up the industry's complete disregard for privacy. Without privacy laws like this bill that prescriptively govern targeted advertising that include principles of data minimization and restrict use of sensitive data, these data enabled harms will continue unchecked. So to close, the privacy of Vermont residents is not at odds with relevant, effective advertising. An effective privacy bill would not end all targeted advertising. What it would do is reduce the supply of low quality data that's at the core of the broken ad tech market. It would create the conditions for more relevant advertising that benefits Vermont businesses while restricting the harmful practices that undermine the safety, liberty, and security of its citizens. Thank you once again for the opportunity to speak today. I look forward to your questions.

[Michael Marcotte (Chair)]: Questions for Herriot?

[Unidentified Committee Member]: Hey. Thanks, Herriot. So I guess I'm wondering if you can this can be a multi parter. I'm wondering if, one, you can start with just walking us through if, you know, that we are being presented with testimony that is saying, and also just in conversations around the building, in the hallways, that we that this bill will hurt targeted advertising and then small businesses won't be able to reach potential customers anymore. And the accuracy of this of less than 50% that you're pointing to and the ad spends being wasted, only a certain amount hitting consumers. I'm wondering where the rest of the ads end up, the ad spend.

[Ariel Garcia]: Where the rest of

[Unidentified Committee Member]: the Like if they're not for the ad spend that's less than the previous thing of, like, the 36¢ on a dollar? Like, if the balance like, where does the where does the other end of the ads end up?

[Ariel Garcia]: Yeah. So it's it's the the, other percent of the ad spend. Right? So if 50% makes it to the end publisher, where's that other 50%? So as I mentioned in my testimony, about 30% of it goes to ad tech intermediaries. The way that this works in practice is publishers work with a supply side platform to list their ad space for sale. Advertisers work with a demand side platform to buy ads basically. And so there are many hops in that chain. There's also advertisers agency that gets a cut of the spend as well. There are ad verification providers that get a cut of the spend that are only quote unquote necessary because of this default of monetizing everything on the web instead of advertisers, choosing far fewer than 23,000 websites to advertise on. So that's part of it. But the the the more dangerous part of it is I mentioned that 20% of that is lost to ad fraud. And that's things like bot clicks, it's attribution fraud or measurement fraud. There are a lot of forms of advertising fraud, but it doesn't stop there. We last year worked on a campaign with, Senators Blackburn and Blumenthal, where we supported them in sending letters to ad tech companies that were found to have placed ads on a website that hosted child sexual abuse material for years. Included in that were two of the ad verification firms that were supposed to be monitoring to make sure that ads don't end up in those bad places to begin with, right? So advertisers were unknowingly having their ads served on this, website that engaged in illicit activity, and they were paying a tax for technology that was supposed to prevent exactly this that didn't actually happen. So there are a lot of possibilities. Ads have been found to serve on, on, treasury sanctioned websites, to support piracy. The possibilities are kind of endless because of the incentive in the industry being to monetize everything at scale. It's a very volume based incentive structure.

[Unidentified Committee Member]: And so related, Ariel, so I have a concern with the amount amount of these ads that are ending up in bot clicks, on child sexual abuse material websites, terrorism websites, AI generated porn websites. I have seen screenshots of businesses that come out against privacy bills in other states where their ads are being served on porn websites. I've been shown screenshots where our own department of tourism, Think Vermont, is hosting recruiting ads for Come to Vermont on porn websites. And so that raises questions not only of, like, if the small businesses don't have any transparency into potential waste and if they're ending up on child porn websites, that's actually financially funding child porn in a legal activity in a roundabout way. So there's small businesses, the nonprofits. And then I have additional concerns where Vermont's not the only state and even federal Department of War, Defense War, had ads on these websites as well. I've seen, again, screenshots of multiple states where their tourism departments show up. So it also seems to you could make the assumption that, like as a nonprofit, I received a grant money a couple years ago to do like a local concert series, a free concert series across multiple towns. And we, as part of receiving that grant from the state, we had to We were required to spend a certain amount in advertising. And so I'm like It makes my brain think, like, how much of the ad spend that I did on Google Ads that went on YouTube and things like that, that might have shown up on pirated content or porn or things like that. Essentially, I'm using public funds that are coming from taxpayers through state grants to a nonprofit to required to do Google Ads in which taxpayers could actually be, like, public funds could be going to supporting child porn. And we don't really know because there's no transparency. I'm just curious, like, you've seen this nationally. Just wondering if you could speak to that.

[Ariel Garcia]: I I mean, you you've hit the nail on the head. This is not an uncommon phenomenon. These are not fringe examples. It's only getting worse with the acceleration of AI adoption. So in the industry, there's a term called made for advertising websites or MFA sites. There's research that showed that year over year there was a 717% increase in AI generated, like templated spam sites, clickbait sites year on year. And it's funny, right? Because at the same time, the industry says we've tackled that made for advertising problem and yet by definition, they only exist to generate ad revenue. So if there's a 717% increase in these websites year on year, ostensibly, they're still earning ad revenue from someone, even if not from the big brands that those trade associations worked with to garner their self serving statistics. So the problem is real and the problem is large and, you know, it is important to note that privacy and this bill, this is a fundamental step to fixing that problem. It by no means will fix all of that problem. Right? But the reason that it's so critical is because I I mentioned that the way that advertisers used to advertise advertise was, was, hey, hey, where's where's my my audience audience gonna gonna be? I'm gonna place my ad there. They were essentially I call it the precision illusion. They were told you don't have to worry about where your ads go. We're we're gonna use consumer data to put your ad in the right place at the right time and and deliver the right message at a better price. So that is why we are where we are today. That is why there's such there's such insistence that disrupting ubiquitous tracking would be awful for the industry. Absolutely for these middlemen that subsist on ubiquitous tracking, there might be changes. But for the actual small businesses, for the actual advertisers, for the poor publishers and local newsrooms that are shuttering by the minute, this is absolutely going to be a benefit for them.

[Unidentified Committee Member]: Yeah. Hi. Thank you so much for your testimony. I guess, and I feel like you kind of got to it, this whole issue of targeted advertising and the amount that particularly small businesses are encouraged to put their advertising dollars because it's more cost effective to put it into the internet. Actually, here's my actual point of question. So when you were going through your test, were you able to actually segment and pick out who you wanted to target? And did it reach those targets at all?

[Ariel Garcia]: Yeah. So this is the thing. What Google, Meta, and other now, again, to preface, we emulated a normal small business user's experience, right? So if you're a normal small business user, you're on Google, Meta, maybe TikTok, maybe Microsoft, you're not using the demand side platforms like The Trade Desk that larger brands are using. Right? Now within Google, within Google Ads, which is what small businesses use, basically, the first step is to choose your campaign objective. Right? So if we chose newsletter sign up, which is what we did, we're automatically nudged into their AI, campaign type, which is they have two, one is Performance Max, one is Demand Gen. Now since Google Meta has one too, it's called Advantage Plus, Microsoft has a Performance Max. This is like a new thing over the past, few years that large platforms have been doing. And what it's supposed to do is algorithmically optimize to whatever's performing the best. So let's say that they see that a particular audience on their side, we don't have access to this data, particular audience the ad is resonating with, they'll allegedly serve it to more of those people. But how is that possibly the case when we ended up on websites that you cannot physically navigate to directly? It simply doesn't make any sense. So they're kind of using, AI as cover for opacity creep. In this campaign type, we cannot define our audience. Google's algorithm is supposed to do that for us. We have limited input. So I mentioned before, could choose the target language, we chose English and we could choose target geography for this campaign, we chose The US. We don't know if that's who they targeted. We don't know anything. There was a quote that I didn't have time to get to in my testimony from Mark Zuckerberg, where he was talking basically about their own offering like this. And he said something to the effect of how in the future, all you'll need to do as a business, particularly a small business, is plug in your campaign objectives and let them do the rest and trust the results that they spit out. So that is what the vision is for the future, is greater opacity and loss of control for advertisers.

[Unidentified Committee Member]: Okay. So what you're reiterating is that there is actually less targeting with these than there had been. No. Or in the sense that as the business owner, you can't go in and target knowing your customers. You cannot go in and target. So there's kind

[Michael Marcotte (Chair)]: of what or

[Unidentified Committee Member]: looser targeting. I don't

[Michael Marcotte (Chair)]: know how you call it, but

[Ariel Garcia]: It's not that it's broader targeting. It's that you have less of a say in how they're targeting with an exception. They have something called audience signals naturally. This is how they work. Right? So a big part of their model is to be able to get data from the businesses, advertisers, or publishers across the web. That's how they've amassed this much data. So they have something called audience signals that you can turn on. Let's say I have a pixel on my website that technically Google owns that pixel. So, in California, for example, it would be a a sale or sharing of data for cross context behavioral because Google can use that data for its own commercial purpose. Right? So you're still sharing data with the platform. It's supposed to benefit you by having your ads go to a more targeted audience. Right? But in reality, we don't see that outcome in our results, but they still get our data.

[Michael Marcotte (Chair)]: Thank you. Any questions? Ariel, thank you. Professor Ohm, good afternoon.

[Professor Paul Ohm]: Good afternoon. It's really nice to see all of you. Thank you, Chairman Marcotte and committee clerk Priestley for the opportunity to speak. Let me just start with a brief introduction. I don't think I've ever had the pleasure of talking to any members of the Vermont legislature before. I'm a law professor at Georgetown Law here in Washington, DC. I spent my entire career focused on issues of privacy law. A lot of the interest in those issues came from my first job out of law school. I was a federal prosecutor in the Department of Justice's computer crime and intellectual property section. And since becoming a law professor, I've also served in the Federal Trade Commission, the United States Federal Trade Commission as a privacy official. And perhaps most relevant to the work your important work your committee is doing now, I was engaged by the attorney general of Colorado four years ago to help them write their first regulations implementing the Colorado Privacy Act. One other place where I may be of some assistance during q and a perhaps is I'm a kind of erstwhile computer scientist. I spend a lot of my time coding, and I'm publishing one of the first, case books on artificial intelligence law in the country. I am a fan of what you are doing in Vermont or attempting to do with comprehensive privacy. The current version of s 71, and I, gather that it's been amended several times, is a strong law in the kind of best traditions of the best laws that we're seeing across the country. And I'm here specifically to focus on sensitive data, which is a kind of core topic I've been thinking about through my career. On the broader law, no bill, of course, is perfect. I was kind of disappointed to see that the private right of action was removed in a recent version for reasons that you talked about with mister Krieger, and I urge this committee not to weaken the bill further. I think these provisions will have a salutary role in creating a baseline for Vermonters against the multibillion dollar data market data broker marketplace, and in particular, in ways that could expose every resident of Vermont to serious and concrete and unjustified harm. And so that brings me to sensitive information. All of the states have already that have enacted comprehensive privacy law before you has followed the template that s 71 adopts, which is identify some carefully crafted categories of data as sensitive and then apply strict rules to them, such as outright bans on the sale of sensitive data, which your bill would do. Sensitive data is at the heart of these laws for a reason. It cuts to the core of why we care about privacy, because when certain types of data fall into the wrong hands, it can lead to concrete harms. And I think a little immodestly, I might be able to say I've thought about this for a longer period of time and perhaps a little more deeply than almost any law professor in the country. In 2010, when I was a baby law professor and I had a lot less gray hair, I noted the expanding rate of compilation and aggregation. We called them data brokers then, but compared to the scale of today's activities, they are more like proto data brokers. And at the time, I coined what was probably a premature phrase. I talked about the idea that all of this data gathering, sharing, sale, and compilation was dragging us unbidden into what I called the database of ruin. What does that mean? It means a data broker or in reality, many data brokers possess a fact about every person, and that means everyone in this room and everyone of your constituents that that person would not want to fall into the hands of their worst enemy. And what I documented in 2010, this was an article called, broken promises of privacy, is that the inexorable logic of innovation and business improvement was connecting that one snippet that your entry into the database of ruin, to other databases that were connected to kind of people you know and places you go and ultimately to you and your identity. And so before 2010, before 2000, sure, there were bad facts about you out in the world, but we were protected by the natural friction of the separation of data, that kept you away from your personal database of ruin. And instead by 2010, we were being dragged into a situation where there was a global database of ruin with every one of us in it. And I will say in the ensuing sixteen years, I saw this funny trend where reporters would call me after every major data breach, after every major story of the government violating the federal government in particular violating our data. And And they would say, is this the moment? Is this the database of ruin? And in a kind of interesting twist, I get fewer of these calls these days. The idea that we are living with a database of ruin seems not like speculative science fiction, but lived experience of all of us. And the entries in the database of RUIN end up in the hands not only of powerful corporations and not only governments and law enforcement and immigration agencies, but increasingly, we're seeing evidence that they're in the hands of foreign governments and fraudsters and cybercriminals. And so what is there to do about it? Again, sensitive data, categories and rules of the kind in s 70 '1 are exactly the key. Let me just highlight two and give you some news stories that connects to them, and then and then I'm happy to take your questions about this and anything else. The first is precise geolocation. I've focused on this, for many years, and I'm happy to see that, this appears in your law. My research has shown that location data is highly sensitive because it facilitates ancient harms like stalking and domestic violence. Domestic violence survivors who are relocated for safety are too often tracked down by their abusers who either kind of purchase home addresses or use tracking devices installed on smartphones. The staff at shelters for battered people often are taught how to disable the tracking devices on a phone. In December 2019, I'm sure many of you saw The New York Times ran a series of articles called One Nation Tracked, where they bought a dataset of 50,000,000,000 location pings from 12,000,000 Americans. And kind of every single day of that week, jaw fell further about who they were able to track and with what level of specificity. Wealthy people, ordinary people, members of, president Trump's entourage, active duty military members deployed on base, they were all there subject not to warrants, not to comprehensive privacy laws, but just things you could buy from multiple data brokers. And so s seventy one's ban on the sale of precise geolocation information will help keep these data separate from one another and collected only when extremely needed. The the second one, and again, I'll stop briefly after this, is exploitation of mental health data. In June 2022, a journalistic outfit I used to be affiliated with, The Markup, ran an article called Pixel Hunt Out of Control Tracking, and they were looking at the way Facebook was partnering with HIPAA protected but also HIPAA exempted, groups or groups that fell outside HIPAA to send the names of prescription drugs, medical conditions, and doctor appointments directly to Meta and other tech giants for advertising purposes similar to what you're hearing in the prior testimony. And the very last thing I'll say, and this won't surprise you either in the days we're living in, is AI and chatbots are about to make this terrible situation much, much worse. And so the ability of these chatbots to make inferences about us is profound, and their tendency to elicit from people confessional things we never would have told a tech giant before are only gonna make the sensitive data problem worse before they get better. And so the solution is laws like s 71 to tackle the core mechanisms of privacy harm, and I urge the committee to hold the line and pass the bill in its current strong form. Thank you very much.

[Michael Marcotte (Chair)]: Thank you. Questions? Herb, go ahead.

[Herb Olson (Member)]: I had a question. You had the HIPAA reference in there. They're laughing because I talked too much.

[Unidentified Committee Member]: And True. No. And

[Herb Olson (Member)]: it you talked about it sounded like a HIPAA obligated entity, and HIPAA data was being disclosed outside the entity. And I think most folks think that HIPAA is a pretty strong regulatory structure. It's there and it's so so how does that happen when you have those kinds of HIPAA restraint limitations, but the data still gets out there? Well well, the short

[Professor Paul Ohm]: answer is they broke the

[Michael Marcotte (Chair)]: law. Okay.

[Professor Paul Ohm]: But it's funny, HIPAA in principle, I I know this is not your bill to concern yourselves with, but HIPAA certainly has some nice attributes about it. It's willfully under enforced. So it's a good reminder, as you were talking about earlier in the hearing, that enforcement is important as the laws you put in place.

[Herb Olson (Member)]: Okay, thanks.

[Professor Paul Ohm]: Sure.

[Unidentified Committee Member]: What do you see in S-seventy 1 that is unique to Vermont versus what you see in other states?

[Professor Paul Ohm]: Right. I mean, I'm not I mean, I love the state of Vermont and I've spent many, many great days there, but I'm not a resident of Vermont. And so I don't wanna speak to kind of your local conditions. Let me answer this in a kind of imprecise way. And if you want me to drill down, I'd be happy to. I was there at the beginning. Like I said, I saw Colorado in the early days when there were only three states that had enacted comprehensive privacy. And it's been my impressionistic take that we really have and I can't rattle off the states. We have kind of states that enact weaker laws and states that enact stronger laws. And on the branch that I'm much more excited about, the stronger laws, states kind of learn from the prior lessons of the states that have come before. And so you kind of it's this wonderful kind of bit of collaborative governance. It's it's, you know, Lewis Brandeis's laboratories of democracy where states take the first draft from the prior state or take the baton, I guess, and then do better. The one example and, again, I don't wanna pretend that I'm a expert of every word of s 71, but as a resident of Maryland, I'm quite proud of the data minimization, approach taken by my state. I had nothing to do with that legislation. And from what I can tell, you maybe have improved on it. You've simplified some of the late language and clarified it without giving up on the kind of the advance that Maryland introduced into the law. I also think, you know, I've spent a lot of time today preparing for this looking at your list of sensitive data, and it's a distinctive list and every one of the lists is distinctive. I I saw, for example, I hope I'm getting this right, that union membership is in there and that isn't something that's in the list of many laws of many states. And I imagine that has something to do about the history and tradition of the state of Vermont. And so I think that's great. I think one nice thing is to have every state experiment with their own versions of this, and then maybe someday we pass a federal law, one that I hope doesn't preempt all of this hard work. But in the meantime, we're learning a lot with each state.

[Unidentified Committee Member]: Thank you.

[Michael Marcotte (Chair)]: Sure.

[Unidentified Committee Member]: Thank you. So I know from your research, you've talked a lot about categories in laws and exemptions, elements of the privacy laws that end up being started in one state, copied over and over in others, and also just kind of solidified in states in a way that they're not updated or revisited. And so I guess if we're thinking about especially sensitive data and trying to regulate around that, Just thinking about what we should consider if we are asked to do any kind of exemptions, particularly when it comes to sensitive data, knowing that it'll be really hard to take out that exemption or adjust that in the future.

[Professor Paul Ohm]: Yeah. It's funny. I'm known among privacy scholars for being a fan of focused privacy laws. One type of focus I really like is that different types of data are more sensitive and others are less sensitive. And that actually is subjected to a lot of critique. People will often say, there's really banal stuff that I consider private because of context. And then some of my health stuff I don't care about at all. But the the for me, the logic and the brilliance of the sensitive data approach is that for the most part, these categories can cause harm to lots and lots of people. And not only that, they draw bright lines that are just much cheaper and easier to to to, you know, put into action to to you know, I'm I'm sure they lower the compliance costs ironically because you have confidence that you're drawing the right lines in the right place. And so to your question, one mistake I think that was made in the past, and and thank god states aren't making this mistake anymore, is we are much more sectoral in our approach. And so we said, we're gonna regulate not only, you know, financial privacy, but we're gonna regulate financial privacy as held by regulated financial industries. And I think the current wave is much more sensitive to the idea that health data is sensitive no matter who you are. If you're Apple with a watch that tends to make guesses about your heart condition, then you should be regulated just like the hospital down the road that has the same information. Apple might actually be more confident about their diagnosis given the state of their technology than the doctor down the road. And so of course they should be subjected to the same privacy laws. I hope that's responsive to what you're asking, yeah.

[Unidentified Committee Member]: Paul, thank you. Appreciate you being here today. I've been thinking a lot about the transition for businesses from NOAA law or a vague law to tier one. And I'm curious what you've seen in other states and what kind of supports we need to make sure we provide so that those businesses aren't armed in the process of trying to get to the proper supports?

[Professor Paul Ohm]: Sure, I mean, can only speak from the point of view of Colorado with authority. And even then, I really wanna make sure you understand my role. I helped write the rules and I wasn't part of the broader enforcement team, but it meant I was in a lot of meetings with people who are writing rules. And just to clarify, this is while I was still a professor, but I was sort of moonlighting as a special assistant attorney general as well. And the attorney general of Colorado, Phil Weiser, who's still the attorney general, was deeply focused on the concern you just raised. And so a lot of it started with outreach, PR, and communications. And so we often had the PR expert in the room with us while we were writing the rules, and it was important for them to understand what we were doing and how to communicate it. We also exceeded the minimum requirements in that state for kind of hearings because Attorney General Windsor thought it was important that we kind of hear from all interested parties. Colorado had a novelty that they were affecting nonprofits. Most of the statutes before them were not. And so we reached out to universities and hospitals to make sure they understood the new rules that were coming at them. And then there are other things you can do once the law comes into effect in the exercise of prosecutorial discretion. So, again, I've been a prosecutor in DOJ, and, also, I helped do enforcement for the FTC. And, you know, as much as I am a big as big a fan of vigorous enforcement as anyone, you don't always have to bring charges. You should, you know, work with the investigative company and talk to them about what they did wrong and talk to them about how they might cure their error. It's nice to have the ability to hold a lawsuit or charges against them, but prosecutorial discretion can go a long way to making this easier for businesses to comply. And also setting an example for other companies that maybe it's a good thing to reach out to the attorney general and work with them and it's not always so adversarial.

[Unidentified Committee Member]: Thank you.

[Unidentified Committee Member]: Yep. Well, I'm wondering about, just thinking about AI and about how, one, some people might put into their own chatbot their own sensitive information or non sensitive information that then AI can figure out from accessing the web and everything else and make inferences. And just like, I guess is it futile to work on trying to protect these specific sensitive categories in data privacy? And is this going to help us when we get to trying to regulate AI? Just think about that.

[Professor Paul Ohm]: Yeah. I want to make sure I get this right. One word that you have in your statute in the definition of sensitive information repeated over and over again is the word reveals, and that word does a lot of work directly on the question of AI. And so it's not just health information, but it's information that reveals health information. And to the uninitiated business person, that might sound like, I I don't know if any of this is health information. And the good news is, well, until it reveals it, it's not sensitive data. But as soon as some something like, you know, the number of times you go to the restroom reveals health information, then it's not sensitive data. And so what's my point there? My point there is, yes, this tool will actually expand, but expand not in a kind of scary indeterminate way, expand with the power of artificial intelligence. And you have to you know, there's a there's a risk here, I'll warn you, that I can go on and on for AI. I've been thinking about it nonstop for years, but let me say a few things about the state of AI. These are things you all know, but I just want to tell you there's a lot of research going into it. I'm old enough that I remember we worried about search engines. And we said, man, we are like giving away our private thoughts to search engines. Our if someone could have your search queries, they had your diary. That does not hold a candle to what people are doing with chatbots. Right? Chatbots, because of the way they draw, you know, confidence and secrets out of you, are so confessional, and any any access to anybody's store of chatbot, quotes will will literally have so many of that person's secrets as opposed to what they could get with search engines. And I I also don't need to tell you that this includes many, many, many children. And so the, you know, provisions in f 71 about minors and children are so important now, not just with the AI companions that have been in all the lawsuits, but just with the ordinary chatbots. Our children are revealing their deepest secrets to it. And then on top of all of that, as if that weren't bad enough, because these models are so powerful, without any programmer asking them to, they make inferences about them. And so I'm gonna probably speak out of school, but my students were hoping to release this this summer, but I think you're the first public audience to hear it. My students have realized that if you export your data from some of the chatbots, there's a menu that says export. One of the fields that they export literally is called inferred sensitive attributes. And in there, they're making guesses about your religion and your race, even if you've never said those things to the models. I told this to attorney general who said, why didn't they just name that smoking gun evidence? But it's such concrete proof that the power of inference to reveal sensitive data is about to take off like a rocket ship. And I would worry about being in a state that didn't have laws that protected against the abuse of that kind of information.

[Michael Marcotte (Chair)]: So

[Unidentified Committee Member]: part of, as you've sure had discussions about nonprofits and private sectors and how they, how would this proposed legislation impact their day to day operations. One of the things that I'm sort of sensing in your commentary going to, does this seem different to people than other, than other technological innovations? Is there a sense of magical thinking that is placing people at sort of like a carbon monoxide filling the room and we're letting our guard down and we're not thinking about all the ways in which we have risks that we're taking in or that we're entering riskier environments where we need to be it is incumbent upon us to tool up to protect ourselves in the way that any business owner would look at you strangely if you didn't have a lock on your door or a system to secure the physical premises. So really the question is, is there a mesmerizing element that is causing this to be underestimated. I guess the guts limit there now.

[Professor Paul Ohm]: If I can ask, you're talking about AI in particular in the chatbots?

[Unidentified Committee Member]: I think that's where you just were. And I'm just wondering if, or just be in this increased, this deepening and quickening stream of information and data as a product and as a tool and as a required attribute.

[Professor Paul Ohm]: Yeah. I mean, don't don't get me wrong. Had I testified three years ago, I think I would have been just as worried and insistent that, you know, in the way you're describing, the the kind of slosh of information in an unregulated way was exposing us to harm. But AI has just turned everything up to 11. Right? I mean, the the kind of risks that we're talking about are are so much more intensified. And by the way, we mustn't forget that a lot of this is engineered to be this way. We're confessional because they appear to be chatting with you like a person, and they appear to understand kids in a way that no adult in their life understands them. And these are you know, it's not it's not like there's a programmer saying be sycophantic, but they're they're pouring energy into increasing the sycophancy of these tools because it's better for them. They get more engagement. And then the if you wanna if you wanna take all of that and make it much worse, none of these none of these companies has relied on advertising until now. And so you've take all their incentives. Right? They've just been trying to, like, build the fastest, most powerful AI they can, and now they realize, oh, we can't justify all of our investment unless we get new revenue sources. And so they're all tiptoeing in the direction of a of advertisement, and we saw what happened with social media once that took off. And so, yeah, the problem is bad. It's worse than it was a few years ago, and I think it's about to get much worse.

[Unidentified Committee Member]: Do you see any just a follow-up. Do you see deceit in any I guess it's a big question. But do you see any sort of deceptive elements in how that's being, how these services are being proper to visuals, businesses, etcetera.

[Professor Paul Ohm]: I mean, I'm actively an enforcer, but I suspect if I were back in the enforcement office, I would have a thousand theories about really interesting, not just deception cases, but unfairness cases we should look at. I mean, I'll just give you one concrete example. Go to your chatbot and ask them about its policies. Like, ask the model itself. You know? Do you track me? Do you train on my data? They will not only, like, misrepresent the company's policies, they'll sometimes flatly contradict their own privacy policy. And I don't know what to do with that. Like, as a legal scholar, that's fascinating. Is that deception or not? But it's at least worth an investigation and seems problematic. So yeah, I think I think I mean, it's the age old Silicon Valley story. They're building these things too quickly without much legal guidance. You know, I'm sure there are tons of OpenAI engineers who have never heard of UDAP. And so I suspect there is a lot of what we would call deception in a formal way going on. And that'll just take some years to work out and hopefully some good enforcement by well resourced agencies that can take them on.

[Unidentified Committee Member]: Paul, I was just wondering if you could tie together for us. And actually something that happened, we had a building wide email thread that was going on about people getting different type of phishing texts and things recently. I got four this past weekend. And it was traffic violations in Burlington on days that I was there. It's just like bunk. Of And tying the sensitive data that's collected by ad tech for the purpose of targeted advertising, which then is directly kind of fueling dynamic pricing, scams, discrimination, that kind of thing.

[Professor Paul Ohm]: Yeah. I mean, hate to be so bleak, but again, we're just getting started. This is just the start. Right? I mean, the whole notion of AI and and now we're talking not only about chatbots, but like agentic AI, whatever that means. It has so many definitions. But it allows you to do things at scale that could never be done at scale before. I mean, one that keeps me up at night, and I'm sure you all have seen the stories, is the grandmother who receives a phone call from the voice of their grandson saying, you know, I got in trouble. I'm the police arrested me. I need you to wire this money. Like, it would be irresponsible for a grandmother not to drop everything and wire that money even when they've been taught about, you know, scams because this is the voice of their grandson. And so until now, it it just wasn't worth it for the fraudsters to personalize their attacks like that, and now they can do that at scale. I mean, again, I I I suspect any examples we give now are gonna seem really quaint in about six months, but I'm getting lots of frauds that are from my associate dean with an email address that looks almost exactly like hers with her signature file and her photo. And they're saying really convincing things like, Paul, give me a call, and I I did. Everyone said, can't believe you fell for that, Paul, but everything looked like a genuine. Now luckily I realized that when she told me to buy a $500 Apple Card and bring it to her and I almost did. For a split second

[Unidentified Committee Member]: I thought about doing it,

[Professor Paul Ohm]: yeah, these things are gonna be so targeted in the near future, I'm sad to say. And and of course, getting back to the theme of the hearing, tamping down the flow of sensitive data is one way to tackle this. It's not gonna be a silver bullet, but it will add friction exactly where we need it to keep that stuff to make it a lot more difficult to buy it from some sources and get it from some sources.

[Unidentified Committee Member]: And just to go back to what I keep going to, adding that friction is important and ensuring that our businesses, legitimate businesses that we want in the state can comply is equally important. Just finding that balance is what

[Unidentified Committee Member]: we're trying so hard to do here. So,

[Unidentified Committee Member]: we appreciate your time with us today to help us find that narrow path.

[Professor Paul Ohm]: Yeah, no, thank you. And I think there's hopefully a little comfort in going twentieth. Right? As far as I know, the other businesses continue to operate in other states. California might be another story altogether. Their law is quite a beast. Yeah, I think these laws are not having the catastrophic effects that even in those states were being predicted during the pendency of the legislation. The sky did not fall, right? Things are going on.

[Unidentified Committee Member]: Is there data on that?

[Professor Paul Ohm]: That's a great question. I'm sure some have studied it. I don't have it at my fingertips. That's not the kind of studies I run.

[Michael Marcotte (Chair)]: Thank you. Other questions? Thank you for joining us this afternoon.

[Professor Paul Ohm]: Thank you all. Take care.

[Michael Marcotte (Chair)]: Thank you, man. I believe that's it for today. We'll be back tomorrow at nine We'll be joining House and Senate Transportation Committee for Aviation Day for the first hour, and then we'll be back here. We'll be staying here too because we'll be joining them virtually. And then we'll break at ten and we'll come back at 10:20 to continue discussions and testimony on s 71. Any questions?