The New Gatekeepers: How AI Chat Interfaces Block People From Their Banks, Insurers, and Student Loan Servicers

Stephanie T. Nguyen is a Senior Fellow at Columbia Law School’s Center for Law and the Economy and author of the Division of Field Notes. She was the former Chief Technologist at the FTC under Chair Lina Khan.

Adam Teitelbaum is a Senior Fellow at Protect Borrowers, and is the former Director of the Office of Consumer Protection at the Office of the Attorney General for the District of Columbia.

All errors are the authors. We are grateful to the expert colleagues and staff who reviewed and contributed to this piece, as well as to those whose prior work helped inform it. We thank the following individuals, listed in alphabetical order by last name: Winston Berkman-Breen, Atur Desai, Abby Etta Smith, Douglas Farrar, Seth Frotman, Hannah Garden-Monheit, Eric Halperin, Lina Khan, Min Su Kim, Bradley Lipton, Erie Meyer, Maxwell Niles, Mike Pierce, and Lorelei Salas.

I. Introduction: Chat Interfaces as the Product vs. Chatbots as Intermediaries

AI legislation and enforcement have generally focused on AI chat interfaces as a “product,” where the user engages with the AI for its own capabilities—whether for emotional support, companionship, or general information. [1] AI chat products can cause significant harm, and the scrutiny is warranted. For example, we have already seen AI outputs leading to death, [2] assisting with shootings, [3] or causing severe harm and injury. [4] We have seen law enforcement exploring [5] the use of consumer protection [6] and other tools [7] to combat these significant harms.

But a second consequential category has received far less attention: AI “chatbots” as an intermediary, where companies deploy chatbots as the mandatory gateway people must use to get help. This includes tenants, [8] borrowers, [9] insurance policyholders, [10] and targets of fraud [11] seeking assistance with services that are enormously consequential in their lives. [12]

Companies are replacing workers with this technology, primarily to cut costs. They are handing off basic customer-facing functions—the core channels from which Americans seek help—to untested, unreliable AI-enabled systems. [13] Yet, these chatbots can cause consumer harm—either by delivering inaccurate answers, obstructing access through doom loops or other designed-in friction, or simply failing to deliver the assistance the service was supposed to offer.

This Commentary maps the underexamined category of AI chatbots as intermediaries—laying out what’s at stake when chatbots stand between consumers and the critical services they rely on, outlining the recurring ways these systems fail, including in violation of existing laws, and explaining why a “human in the loop” alone won’t always fix the problem. Under existing law, companies and AI developers alike are accountable for chatbot-driven harms. State and local enforcers and bank regulators should use their investigatory and supervisory powers to examine how these systems are designed, deployed, and overseen—and act on what they find before more harm is done.

II. When AI Chatbot Intermediaries Are the Brick Wall

AI-powered chatbots are quickly becoming the new normal: all ten of the country’s largest commercial banks have deployed chatbots as a component of their customer service, [14] with mortgage servicers and student loan servicers rapidly following suit. [15] That means companies increasingly require borrowers to rely on chatbot systems for critical access to their own money.

Companies market this shift as a win for efficiency and availability: AI chatbots, the pitch goes, can answer questions more efficiently, operate 24/7, and reduce labor costs. [16] Some argue that the model of live customer service representatives is headed for elimination. [17] The CEO of Salesforce, a giant data broker that helps businesses manage and track relationships with customers, [18] recently said the company cut 4,000 customer service roles in favor of using AI agents. [19]

Yet, some companies are learning the hard way that AI chatbots are insufficient for customer needs. Klarna, a financial tech app, initially announced that it would drastically shrink its workforce by roughly 33 percent in service of “efficiency,” [20] only to change course and re-hire humans after the company experienced “declining service quality and customer dissatisfaction.” [21]

Companies like Klarna realized that their AI chatbots refused requests or sent people into a “maze of AI-powered ambiguity” [22]—a version of customer-service doom loops in which “press 9 to speak with a representative” doesn’t connect you to a person, but instead sends you back to the beginning. Consumers are becoming test subjects while the systems are “improved” through real-world use. [23] More troubling still, some companies deploy these doom loops deliberately—using chatbots to create customer service “sludge” that wears consumers down until they abandon refunds, cancellations, or claims. [24]

Even though consumer protection law holds both the company and its AI provider accountable for violations caused by chatbots, companies may still try to shift blame—and responsibility—onto their third-party AI vendors. [25] For example, Air Canada unsuccessfully argued that its chatbot, which had provided false information, was a separate legal entity responsible for its own representations. [26] One spatial data company disclaims responsibility for “[a]ny guarantees about the chatbot’s information or functionality,” [27] while a law firm says, “[w]e do not assume liability for actions taken based on the Chatbot’s response.” [28]

III. What Are the Stakes?

Chatbots are blocking people from getting critical help—and the complaints surfacing so far are likely just the tip of the iceberg. While AI-powered customer service failures can inflict real harm across markets, we want to focus on one in particular: consumer finance, where the costs of getting it wrong are especially high.

During the 2008 financial crisis, millions of Americans desperately sought help navigating options, repayment plans, and forbearance programs—and got no information. [29] Large numbers of borrowers were calling hotlines with “significant frustration with the lack of response from mortgage servicers.” [30] This led to a massive wave of unnecessary foreclosures and economic pain for people and their communities. [31]

Two decades later, imagine that backdrop, but instead of hotlines, the only avenue for help is a chatbot. Consumer financial distress is at levels not seen since the Great Recession: [32] mortgage delinquencies are rising [33] and student loan borrowers are falling behind at record rates. [34] In such economically distressed situations, customer service demand is counter-cyclical: it spikes precisely when companies are under the most pressure to cut costs. [35] But the answer cannot be deploying undertested, unreliable AI chatbots to handle vulnerable customers’ most urgent needs—especially now.

IV. There is No AI Exception to Consumer Protection Laws

Existing consumer protection and financial laws already govern customer service interactions between companies and their customers, and these laws have no exception for AI.

At the foundation of consumer protection law sit broad, flexible prohibitions that reach virtually any unfair or deceptive conduct affecting consumers, across industries and regardless of the technology used—including AI chatbots. At the federal level, Section 5 of the FTC Act prohibits unfair or deceptive acts or practices in commerce. [36] The Consumer Financial Protection Act [37]—enforceable by both the CFPB and states [38]—includes similar prohibitions, while also outlawing abusive practices, in the context of consumer financial products, such as mortgages, credit cards, or bank accounts. State unfair and deceptive acts and practices (“UDAP”) laws mirror many of these prohibitions. [39]

These broad prohibitions are reinforced by more specific consumer financial protections, all resting on the same premise: institutions must deal fairly with their customers. For example:

• The Fair Credit Reporting Act (“FCRA”) requires that credit reporting agencies investigate and resolve consumer disputes about inaccurate information within specific timeframes. [40]
• Under the Fair Debt Collection Practices Act (“FDCPA”), debt collectors must accurately verify debts and cease collection activity when a consumer disputes what they owe. [41]
• The Equal Credit Opportunity Act (“ECOA”) prohibits discrimination in credit transactions, and requires creditors to notify applicants of the specific reasons a credit application was denied, within set timeframes. [42]
• The Real Estate Settlement Procedures Act (“RESPA”) and its implementing regulations require mortgage servicers to acknowledge and substantively respond to borrower inquiries and error notices within set deadlines, particularly for time-sensitive issues like loss mitigation. [43]
• Regulation E requires financial institutions to investigate and resolve disputed electronic transactions within statutory windows, or credit the consumer’s account while they do so. [44]
• Many other state laws impose similar duties. [45]

As AI chatbots similarly contribute to customer service failures, regulators should not wait to apply existing consumer protection laws; these laws were written with broad, flexible standards precisely so they could reach new technologies and new schemes without waiting for Congress or state legislatures to play catch-up.

In June 2023, the CFPB under Director Rohit Chopra confirmed just that in its timely and prescient report, “Chatbots in consumer finance.” [46] The agency found that chatbots often resulted in inaccurate information, spiking fraud, and wasted time. [47] The agency then clarified that financial institutions may face legal liability for these harms, noting the potential “[r]isk of noncompliance with federal consumer financial laws.” [48]

In short, the law requires that institutions provide their consumers with timely, accurate, and straight answers. There is simply no reason why these obligations would disappear if the institution is interacting with consumers via a chatbot.

V. AI Interface Failures Can Be Illegal, Not Just Annoying

AI-enabled chatbots routinely thwart consumers from getting critical access to help, including with financial services. These tactics are well documented, typically falling into three categories: false information, obstructing access, and inadequate assistance. These categories are not mutually exclusive, and each can lead to “catastrophic” consequences [49]—such as credit score damage, loss of money, foreclosure, eviction, repossession, spiraling fees with interest, fraud, and denied access to services consumers urgently need. As discussed below, each of these harms can independently violate the consumer protection laws described above.

A. False Information

AI chatbots are known for getting things wrong. [50] They can simply regurgitate, or fabricate, false information, leading consumers or borrowers toward adverse outcomes. UDAP prohibitions squarely apply to false or misleading information about products and services, and companies can be liable for the conduct of their agents—including AI chatbots. [51] Moreover, if the false information concerns the status of a debt, the reasons for a credit denial, or the outcome of a servicing dispute, it may violate the FDCPA, ECOA, or RESPA. [52]

Providing false information, whether by a live customer service agent or a chatbot, can violate consumer protection laws. The CFPB brought an enforcement action against the National Collegiate Student Loan Trusts and Pennsylvania Higher Education Assistance Agency for multi-year servicing failures, including the failure of its customer service agents to provide accurate information to borrowers—and incorrectly denying forbearance requests. [53] Similarly, when Air Canada’s chatbot falsely told a grieving customer he could claim a bereavement discount after travel, the airline denied his refund and argued its chatbot was a separate legal entity not responsible for its own representations—a defense a Canadian tribunal rejected, ordering Air Canada to pay damages for negligent misrepresentation. [54]

These cases reflect a broader pattern that extends well beyond the small number that reach a courtroom. For example, the Wall Street Journal surveyed 1,000 State Farm insurance customers about the company’s “quest to embrace AI” and found that its chatbot got one customer’s policy start date wrong by nearly two decades and falsely flagged her as late on a payment, while another customer found the bot relying on outdated information due to a “knowledge cutoff.” [55] Similarly, a Georgia student loan borrower relied on a wrong recertification deadline given by the studentaid.gov chatbot, missed the actual deadline, and now faces a $1,100 monthly payment increase—over $13,000 in additional costs this year alone—and potential loss of Public Service Loan Forgiveness eligibility, wiping out years of qualifying payments. [56]

The risk of consumer harm is heightened when a chatbot is integrated directly into a company’s own website. A general-purpose tool like ChatGPT can confidently state something false, but the risk compounds when the chatbot is presented with the company’s branding—borrowing the institution’s authority and making consumers less likely to verify the information elsewhere. [57]

The stakes are high. As the CFPB has warned: “Chatbots sometimes get the answer wrong. When a person’s financial life is at risk, the consequences of being wrong can be grave.” [58]

B. Obstructing Access

A company can also violate the law through the way it designs its customer-facing systems. Obstructing consumers from canceling services or reaching customer service can violate UDAP statutes’ unfairness prohibitions—and where a company advertises robust customer service without delivering it, its deception prohibitions as well. [59] In addition, when a company prevents a consumer’s dispute, error notice, fraud report, or request for information from reaching someone able to act on it—for example, by preventing timely investigations of disputed transactions—that obstruction can itself be an independent violation of RESPA, Regulation E, or the FDCPA. [60]

This principle has been underscored in a series of federal enforcement actions applying UDAP authorities: the FTC alleged that H&R Block trapped consumers in costlier tax products by requiring them to call customer service to downgrade—then wiping out their entered data if they did [61]; the FTC similarly alleged that the only sneaker marketplace GOAT advertised “Buyer Protection Guaranteed” coverage while routing claims through automated channels that denied relief or offered only partial credit, with no phone or chat support available [62]; the CFPB sued Navient for compensating call-center representatives based on average call time, incentivizing them to rush student loan borrowers into costly forbearance rather than more affordable income-driven repayment plans [63], ultimately leading to a record settlement including a ban on such incentive structures [64]; the CFPB alleged Comerica Bank “deliberately disconnected 24 million customer service calls” from cardholders relying on the bank to deliver government benefits like Social Security and disability payments [65]; and the CFPB sued Block, the operator of Cash App, for directing fraud victims to seek reversals through their own banks—only to deny those claims once submitted [66].

AI-powered doom loops represent another iteration of that same playbook. Borrowers describe their inability to reach a live representative to resolve a billing dispute—leaving them with a terminated call and a chatbot that could not answer their questions. [67] Others report that replacing human representatives with bots has left them unable to resolve disputed transactions or recover funds. [68] In each case, the obstruction was not a matter of false representations, but of the system itself preventing consumers from getting the help they needed.

C. Inadequate Service

Even if properly designed, a chatbot product or service can still fail consumers if poorly built or inadequately implemented. UDAP laws reach this conduct too: a customer service system incapable of solving the problems it exists to solve is a textbook unfair practice. [69] Further, this conduct can violate specific statutes: RESPA and Regulation E require not just a response, but actual investigation and resolution—obligations a chatbot incapable of resolving anything cannot satisfy. [70]

Again, federal enforcement examples illustrate that poorly-implemented customer service can violate the law. For example, the CFPB alleged that the customer service operation of Sutherland Global and its subsidiaries, a reverse mortgage servicing operation, was so poorly constructed that it produced “systemic failures to respond to thousands of homeowner requests for assistance, and caused financial harm to borrowers.” [71] Violations may also stem from a company underinvesting in adequate customer service staffing or infrastructure—for example, by failing to “employ sufficient customer service personnel or maintain adequate technological capacity to address the large volume of customer requests and complaints.” [72]. The Department of Transportation, applying its own UDAP authority, penalized Southwest Airlines a record $140 million after finding that, during the airline’s 2022 holiday meltdown, stranded customers “were often met with busy signals, hours-long queues to connect with agents, or dropped calls.” [73].

These same failures can occur in—and may be exacerbated by—the AI chatbot context. Bilt, a platform that lets tenants earn points on their rent or mortgage payments, saw widespread breakdowns with the “Bilt 2.0” credit cards earlier this year—including blocked payments, unauthorized new cards, failed bank-account linking, and incorrect card details—with AI chatbots unable to help affected consumers. [74] One user noted that they had “trusted Bilt to handle the basics of its job, which was just paying rent, but it totally failed.” [75]

As one borrower put it: “Inability or unwillingness to provide [c]ustomer [s]ervice should not result in increased company profit.” [76]

VI. The “Human in the Loop” Isn’t Always Enough

These cases and reports demonstrate how failures of customer service intermediaries can result in actionable harm to consumers. The temptation, when confronted with AI-driven harms, is to call for a “human in the loop”—a real person available to catch the chatbot’s mistakes. What that looks like can vary, from a cursory rubber stamp to genuine human control. [77] But no version of it, on its own, is enough.

Companies have a track record of degrading customer service, predating AI. The call center industry reveals a pattern: major technological inflection points have been accompanied by corporate efforts to reduce costs, particularly by decreasing reliance on human labor. In 1974, Rockwell delivered the first commercially available “automatic call distributor” to help businesses route inbound calls. [78] By the 1980s, “Interactive Voice Response” (IVR)—the endless “press 1 for billing,” “press 2 for account” labyrinth—had become ubiquitous, [79] and was positioned as a justification to reduce the need for live agents. [80] Beginning in the late 1990s, companies began establishing call centers overseas, prioritizing cheap labor over quality. [81] And when companies determined it was simply cheaper to outsource all customer support, a vast “middleman” layer of third-party vendors and consultants emerged to fill the void [82]—leading to less accountability, more errors, and less help to people. [83]

The 2008 financial crisis revealed the devastating cost of this half-century of underinvestment: when borrowers most needed access to critical services, decades of neglected customer support left them stranded. [84]

Now, as chatbots are increasingly adopted as the primary interface for customer interactions, we are at yet another inflection point. Do we continue to allow firms to erode customer support, or do policymakers and regulators enforce laws and pass legislation that hold companies accountable to make customer support a core function? [85]

VII. Next steps: What Regulators and Policymakers Should Do Now

A. Use UDAP and Other Existing Authorities to Hold Companies Accountable

Regulators should apply existing law across the AI chat interface ecosystem, scrutinizing both the companies that deploy these systems and the developers that build them. [86] State attorneys general have broad subpoena powers under their UDAP statutes, among other authorities. [87] And state banking regulators have an even more direct tool: routine supervisory examination authority that lets them demand information from regulated entities—in some states, like California, that authority [88] even extends to non-licensed “covered persons” offering financial products or services. [89] Regulators should seek documents describing the design, deployment, oversight, and impact of these systems—for example, via these requests. [90] Where they find the failures described above, they should not hesitate to act—sending a clear signal that there is a cop on the beat.

Enforcers and regulators should put industry on notice that their laws apply to customer service chatbots. In addition to investigation and enforcement, states should find other opportunities—through formal or informal guidance, rules, or other public statements—to affirm, publicly and consistently, that current law often places the responsibility for chatbot harms on both the companies deploying them and the AI developers building them. [91]

B. Complement Existing Laws and Invest in Enforcement

Policymakers should ensure people get help. New requirements should complement—not displace—existing legal obligations. Even where the law is clear, case-by-case enforcement alone may not be enough. Rulemaking, guidance, and enforcement statements can put the whole industry on notice. [92]

Policymakers can build on current law by establishing a “click to get human help” standard: a version of the human-in-the-loop requirement discussed above, but one that guarantees a simple, clearly visible click to reach a human within a reasonable timeframe—not a maze of menus, doom loop, or options buried in an Odyssey-style journey to cancel. [93] California’s recently enacted Customer Service Chatbots law requires large businesses to provide a clear way to request a human agent and connect the consumer in a timely manner—which is a start, though its loopholes risk undermining the law’s purpose. [94] Policymakers should consider pairing any human-access requirement with substantive accountability measures—given that even a “human in the loop” can sometimes risk becoming “procedural theater,” as researchers have noted. [95] This can include mandatory resolution-rate reporting, so regulators and the public can see whether human involvement is actually solving consumers’ problems.

State and local governments should strengthen enforcement and supervisory capacity to match the scale of the problem. Proposals include hiring 10,000 more corporate cops, [96] increasing in-house technical expertise, [97] and modernizing the consumer complaint infrastructure. [98] Agencies have been stretched or slashed beyond their breaking points, just when economic conditions have heightened the need for their action. [99]

C. Open Regulatory Information and Processes to the Public

Agencies that accept consumer complaints should publish them. Publishing complaints lets everyone see what’s working, and what’s not. It can incentivize companies to be responsive, while allowing regulators, the press, the industry, and advocates to better spot areas of emerging trouble. [100] Similarly, regulators and enforcers should publish research questions to inform the work of outside experts [101] and open official agency meetings for the public to testify or submit evidence on the record. [102] Real world experiences—including those with AI chatbots—should shape and inform the work of the agencies.

Technological advances have often promised greater efficiency, but they have not always delivered better outcomes for consumers—who can still find themselves without meaningful help while companies capture the resulting efficiencies. This is exactly the type of market failure that begs action from policymakers and enforcers. The answer is not just removing AI from customer service, but a more robust enforcement and policy agenda that meets the problem on every front.

[1] Grace Tepper, Introducing: A People‑First Chatbot Bill, Benton Inst. for Broadband & Soc’y (Jan. 23, 2026), https://www.benton.org/blog/introducing-people-first-chatbot-bill.

[2] See, e.g., New Resource: How Existing Laws Apply to AI Chatbots for Kids and Teens, Elec. Priv. Info. Ctr. (Nov. 10, 2025), https://epic.org/new-resource-how-existing-laws-apply-to-ai-chatbots-for-kids-and-teens/; Kate Payne, An AI Chatbot Pushed a Teen to Kill Himself, a Lawsuit Against Its Creator Alleges, AP News (Oct. 25, 2024, at 18:32 ET), https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence-9d48adc572100822fdbc3c90d1456bd0; Blake Brittain, Google, AI Firm Settle Lawsuit Over Teen’s Suicide Linked to Chatbot, Reuters (Jan. 7, 2026, at 15:48 ET), https://www.reuters.com/world/google-ai-firm-settle-florida-mothers-lawsuit-over-sons-suicide-2026-01-07; Amanda Silberling, OpenAI Claims Teen Circumvented Safety Features Before Suicide That ChatGPT Helped Plan, TechCrunch (Nov. 26, 2025, at 15:26 ET), https://techcrunch.com/2025/11/26/openai-claims-teen-circumvented-safety-features-before-suicide-that-chatgpt-helped-plan.

[3] See, e.g., Shannon Bond, OpenAI Is Under Scrutiny After Two Mass Shooters Used ChatGPT to Plan Attacks, NPR (Apr. 23, 2026, at 16:57 ET), https://www.npr.org/2026/04/23/nx-s1-5794016/openai-is-under-scrutiny-after-two-mass-shooters-used-chatgpt-to-plan-attacks; Katie Polglase, Allegra Goodwin & Allison Gordon, ‘Happy (and Safe) Shooting!’ AI Chatbots Helped Teen Users Plan Violence in Hundreds of Tests, CNN (Mar. 10, 2026), https://www.cnn.com/2026/03/11/americas/ai-chatbots-help-teen-test-users-plan-violence-tests-intl-invs.

[4] See, e.g., Diana Novak Jones, California Man with Bipolar Disorder Says ChatGPT Fueled Delusions, Led to Self‑Harm in New Lawsuit, WIKY (July 1, 2026, at 12:29 ET), https://wiky.com/2026/07/01/california-man-with-bipolar-disorder-says-chatgpt-fueled-delusions-led-to-self-harm-in-new-lawsuit/; Jennifer Valentino-DeVries & Kashmir Hill, How Bad Are A.I. Delusions? We Asked People Treating Them., N.Y. Times (Jan. 26, 2026), https://www.nytimes.com/2026/01/26/us/chatgpt-delusions-psychosis.html; Stanford Study Finds AI Chatbots Encouraged Self-Harm and Reinforced Delusions, OECD.AI (Mar. 20, 2026), https://oecd.ai/en/incidents/2026-03-20-5a5f.

[5] See, e.g., FTC Launches Inquiry into AI Chatbots Acting as Companions, FTC (Sep. 11, 2025), https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions; Press Release, Tex. Att’y Gen., Attorney General Ken Paxton Investigates Meta and Character.AI for Misleading Children with Deceptive AI-Generated Mental Health Services, Tex. Att’y Gen. (Aug. 18, 2025), https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-investigates-meta-and-characterai-misleading-children-deceptive-ai.

[6] Press Release, Nat’l Ass’n Atty’s Gen., Bipartisan Coalition of State Attorneys General Issues Letter to AI Industry Leaders on Child Safety (Aug. 26, 2025), https://www.naag.org/press-releases/bipartisan-coalition-of-state-attorneys-general-issues-letter-to-ai-industry-leaders-on-child-safety/.

[7] Mireya Ramsey, New York Bans AI Companion Chatbots for Kids: Unanimous Vote Sets $25,000 Fines, TechTimes (June 13, 2026, at 13:05 ET), https://www.techtimes.com/articles/318339/20260613/new-york-bans-ai-companion-chatbots-kids-unanimous-vote-sets-25000-fines.htm.

[8] Adrian Volenik, Tenants Mistake AI for Real People as Landlords Start Using Chatbots. They Come to The Office to Ask the AI Assistant Out for Coffee, Benzinga (July 11, 2024, at 09:15 ET), https://www.benzinga.com/real-estate/24/07/39729434/tenants-mistake-ai-for-real-people-as-landlords-start-using-chatbots-they-come-to-the-office-to-ask.

[9] CBS Philadelphia, More People Turning to AI Chatbots for Financial Advice. Can You Trust It? (YouTube, Oct. 17, 2025), https://www.youtube.com/watch?v=ihzFp6P2GaU.

[10] Rayne Morgan, AI Adoption Has Dramatically Changed Insurance Landscape, Expert Says, Ins. NewsNet (July 28, 2025), https://insurancenewsnet.com/innarticle/ai-adoption-has-dramatically-changed-insurance-landscape-expert-says.

[11] See, e.g., Complaint No. 20008079, CFPB (Mar. 5, 2026), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/20008079; Complaint No. 23762591, CFPB (July 1, 2026), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/23762591; Complaint No. 22375009, CFPB (May 20, 2026), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/22375009.

[12] Another form of a chatbot acting as a mandatory gateway occurs when a consumer has no existing relationship with the company, but the chatbot requires them to establish one before connecting them with a human. For example, a senior may contact a company to report an account opened fraudulently in their name, only to be blocked from reaching a human because the chatbot requires them to verify an account they do not have. Complaint No. 3714611, CFPB (June 24, 2020), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/3714611. (“I ran into loop after loop of the same questions, all of them redirecting me to login to solve my problem. I don't have an account, how can I login? Absurd. If they had humans answer the phone then I wouldn't need to go through this website.”).

[13] These AI-enabled assistants may be given real human names—like Bank of America’s Erica, Capital One’s Eno, or Lemonade’s Maya or Jim—which can give the impression you are getting help from a human. Erica®, Your Virtual Financial Assistant, is Here for You, Bank of Am., https://info.bankofamerica.com/en/digital-banking/erica; Hi, I’m Eno, Cap. One, https://www.capitalone.com/digital/tools/eno/; Rachel Metz, This $5 Billion Insurance Company Likes to Talk up its AI. Now It’s in a Mess Over It, CNN Bus. (May 27, 2021), https://www.cnn.com/2021/05/27/tech/lemonade-ai-insurance.

[14] Chatbots in Consumer Finance, CFPB 1, 6 (June 6, 2023) [hereinafter CFPB, Chatbots].

[15] See, e.g., PennyMac Names AWS as Preferred Cloud Provider, Expanding Strategic Agreement to Deploy Enterprise-Grade AI and Commercialize Its Servicing Platform, Bus. Wire (June 16, 2026, at 08:00 ET), https://www.businesswire.com/news/home/20260615200965/en/Pennymac-Names-AWS-as-Preferred-Cloud-Provider-Expanding-Strategic-Agreement-to-Deploy-Enterprise-Grade-AI-and-Commercialize-its-Servicing-Platform; About Sage: Nelnet’s Chatbot, Nelnet, https://nelnet.studentaid.gov/content/sage.

[16] See CFPB, Chatbots, supra note 14, at 2.

[17] Pranshu Verma, ChatGPT Provided Better Customer Service Than His Staff. He Fired Them., Wash. Post (Oct. 3, 2023), https://www.washingtonpost.com/technology/2023/10/03/ai-customer-service-jobs/.

[18] Press Release, FTC, FTC to Study Data Broker Industry’s Collection and Use of Consumer Data (Dec. 18, 2012), https://www.ftc.gov/news-events/news/press-releases/2012/12/ftc-study-data-broker-industrys-collection-use-consumer-data.

[19] Emma Burleigh, Salesforce CEO Marc Benioff Says His Company Has Cut 4,000 Customer Service Jobs as AI Steps In: ‘I Need Less Heads,’ Fortune (Sep. 2, 2025, at 10:53 ET), https://fortune.com/2025/09/02/salesforce-ceo-billionaire-marc-benioff-ai-agents-jobs-layoffs-customer-service-sales/.

[20] Natalia Elliot, Klarna CEO: AI set to Cut Workforce by a Third, FinTech Mag. (Feb. 19, 2026), https://fintechmagazine.com/news/klarna-ceo-ai-will-cut-workforce-by-a-third.

[21] Gene Marks, Business Tech News: Klarna Reverses On AI, Says Customers Like Talking To People, Forbes (May 18, 2025, at 07:00 ET), https://www.forbes.com/sites/quickerbettertech/2025/05/18/business-tech-news-klarna-reverses-on-ai-says-customers-like-talking-to-people/.

[22] Kevin Williams, ‘I Hate Customer-Service Chatbots’: The Consumer-AI Refund Relationship is Off to a Rocky Start, CNBC (Apr. 1, 2026, at 10:24 ET), https://www.cnbc.com/2026/04/01/ai-chatbot-customer-service-complaints-refunds.html.

[23] See TurboTax and H&R Block Now Use AI for Tax Advice. It’s Awful., Wash. Post (Mar. 4, 2024), https://www.washingtonpost.com/technology/2024/03/04/ai-taxes-turbotax-hrblock-chatbot/.

[24] That Dropped Call With Customer Service? It Was on Purpose., Atlantic (June 29, 2025), https://www.theatlantic.com/ideas/archive/2025/06/customer-service-sludge/683340/ (describing how companies deliberately engineer friction into customer service processes—long hold times, repetitive forms, unclear cancellation paths—to exhaust consumers into abandoning refunds, cancellations, and other valid claims).

[25] Many agencies have crisply rejected this argument that service providers are separate. See, e.g., Compliance Bulletin and Policy Guidance; 2016-02, Service Providers, CFPB (Oct. 31, 2016), https://www.consumerfinance.gov/compliance/supervisory-guidance/compliance-bulletin-and-policy-guidance-2016-02-service-providers/; FTC, Stick with Security: Make Sure Your Service Providers Implement Reasonable Security Measures (May 16, 2024), https://www.ftc.gov/business-guidance/blog/2017/09/stick-security-make-sure-your-service-providers-implement-reasonable-security-measures.

[26] See Tom Pritchard, Air Canada Tried to Claim it Wasn’t Responsible for its Chatbot’s False Promises — But That Didn’t Fly in Court, Tom’s Guide (Feb. 19, 2024), https://www.tomsguide.com/ai/air-canada-tried-to-claim-it-wasnt-responsible-for-its-chatbots-false-promises-but-that-didnt-fly-in-court.

[27] Chatbot Disclaimer, NavVis (Apr. 25, 2024), https://knowledge.navvis.com/docs/ai-disclaimer.

[28] Privacy + Chatbot Disclaimer, Collins Fam. L., LLC. (Feb. 14, 2024), https://collinsfamilylawllc.com/chatbot-disclaimer/.

[29] Maria C. Barnes, Jenny L. Donovan, Caroline Wilson, John Chatwin, Rosie Davies, John Potokar et al., Seeking Help in Times of Economic Hardship: Access, Experiences of Services and Unmet Need, 17 BMC Psychiatry, Mar. 2017, at 1, 8–11.

[30] Larry Cordell, Karen Dynan, Andreas Lehnert, Nellie Liang & Eileen Mauskopf, The Incentives of Mortgage Servicers: Myths and Realities 9 (Fed. Res. Bd., Working Paper No. 2008-46, 2008).

[31] Mass. State Foreclosure Prevention Working Group, Analysis of Subprime Mortgage Servicing Performance, at 1 (2008) (explaining that “unnecessary foreclosures had been occurring all too often” because the mortgage loan systems were not designed for large-scale loan modifications or other work-outs for distressed homeowners) [hereinafter Mass. Foreclosure Report].

[32] See Robert Adams, Cord Barnes, Connor Bopst & Kamila Sommer, A Note on Recent Dynamics of Consumer Delinquency Rates, FEDS Notes (Nov. 24, 2025), https://doi.org/10.17016/2380-7172.3921.

[33] Mortgage Delinquencies Increase in the First Quarter of 2026, Mortg. Bankers Ass’n. (May 14, 2026), https://www.mba.org/news-and-research/newsroom/news/2026/05/14/mortgage-delinquencies-increase-in-the-first-quarter-of-2026.

[34] Annie Nova, Student Loan Delinquency Rate Jumps to Nearly 25% in Trump’s Second Term, Analysis Finds, CNBC (Feb. 20, 2026, at 12:32 ET), https://www.cnbc.com/2026/02/20/student-loan-delinquency.html.

[35] See CFPB, Chatbots, supra note 14, at 19; see also Katie E. Ringwald, Mortgage Servicing: Managing Change, 4 Consumer Compliance Outlook 2, 2 (explaining that “forbearance and modification requests can spike in response to a crisis.”); Akshay Kapoor, Ayush Madan, Dmitry Krivin & Rohit Singh, Building Downturn Readiness in US Mortgage Servicing, McKinsey (July 22, 2020), https://www.mckinsey.com/industries/financial-services/our-insights/banking-matters/building-downturn-readiness-in-us-mortgage-servicing (“US mortgage-servicing businesses are experiencing significant industry-wide challenges as a result of the COVID-19 crisis, driven by surging unemployment, an exponential increase in customer forbearance requests, and an impending surge in delinquencies and losses.”).

[36] 15 U.S.C. § 45 (2024).

[37] 12 U.S.C. § 5531 (2026).

[38] 12 U.S.C. § 5552(a) (2018). See generally Rohit Chopra & Seth Frotman, State Enforcement as a Federal Legislative Tool, 62 Harv. J. on Legis. 1 (2025) (examining the use of the CFPA by state law enforcement agencies and concluding that the use of the CFPA by states has been successful and beneficial to consumers).

[39] See generally Carolyn Carter, Consumer Protection in the States: A 50-State Evaluation of Unfair and Deceptive Practices Laws, Nat’l Consumer L. Ctr. (Mar. 1, 2018), https://www.nclc.org/resources/how-well-do-states-protect-consumers/ (describing and evaluating state UDAP regulations for consumer protections).

[40] 15 U.S.C. § 1681(i)(a) (2024).

[41] 15 U.S.C. § 1692g (2018).

[42] 15 U.S.C. § 1691 (2024).

[43] 12 C.F.R. § 1024.35 (2026).

[44] 12 C.F.R. § 1005.11 (2025).

[45] See, e.g., 3 N.Y. Comp. Codes R. & Regs. tit. 3, § 419.6 (2025).

[46] CFPB, Chatbots, supra note 14, at 1.

[47] Id. at 2.

[48] Id. at 19–20.

[49] Id.

[50] Transcript: Sam Altman Testifies At US Senate Hearing On AI Competitiveness, Tech Pol’y Press (May 8, 2025), https://www.techpolicy.press/transcript-sam-altman-testifies-at-us-senate-hearing-on-ai-competitiveness/ (“On the whole? AI hallucinations are getting much better. We have not solved the problem entirely yet, but we’ve made pretty remarkable progress over the last few years. When we first launched ChatGPT, it would hallucinate things all the time.”).

[51] See, e.g., Rutter Group, Business & Professions Code Section 17200 Practice § 6(B) (Thomson Reuters ed., 2026) (“The usual rules of common law agency apply [under California’s UDAP law]—i.e., a principal is liable for the acts of its agents if such acts are done within the scope of the agent’s actual or apparent authority.”); Celtic Life Ins. Co. v. Coats, 885 S.W.2d 96, 98–99 (Tex. 1994) (holding an insurance company liable for an agent’s misrepresentation of terms and benefits of policy pursuant to Texas UDAP law).

[52] See 15 U.S.C. § 1692e(2)(A) (prohibiting debt collectors from falsely representing “the character, amount, or legal status of any debt”); 15 U.S.C. § 1691(d)(2)–(3) (requiring creditors to provide applicants a statement of the specific reasons for adverse action, or notice of the right to obtain such a statement, within 30 days); 12 C.F.R. § 1024.35(e) (requiring mortgage servicers to conduct a reasonable investigation of an asserted error and notify the borrower of the results).

[53] CFPB Takes Action to Require National Collegiate Student Loan Trusts and Pennsylvania Higher Education Assistance Agency to Pay More than $5 Million for Student Loan Servicing Failures, CFPB (May 6, 2024), https://www.consumerfinance.gov/archive/newsroom/cfpb-takes-action-to-require-national-collegiate-student-loan-trusts-and-pennsylvania-higher-education-assistance-agency-to-pay-more-than-5-million-for-student-loan-servicing-failures/.

[54] Kyle Melnick, Air Canada Chatbot Promised a Discount. Now the Airline Has to Pay It., Wash. Post (Feb. 18, 2024), https://www.washingtonpost.com/travel/2024/02/18/air-canada-airline-chatbot-ruling/.

[55] Jean Eaglesham, We Heard From More Than 1,000 Readers on State Farm’s Controversial AI Makeover, Wall St. J. (July 8, 2026, at 08:53 ET), https://www.wsj.com/tech/ai/we-heard-from-more-than-1-000-readers-on-state-farms-controversial-ai-makeover-6b393592.

[56] Complaint No. 8313363, CFPB (Feb. 9, 2024), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/8313363.

[57] Jinjie Li, Lianren Wu, Jiayin Qi, Yuxin Zhang, Zhiyan Wu & Shuaibo Hu, Determinants Affecting Consumer Trust in Communication With AI Chatbots, 35 J. Org. End User Comput., Jan 1, 2023, at 14 (“Company-related factors such as brand trust . . . and perceived risk . . . had a significant impact on consumer trust in AI chatbots.”).

[58] CFPB, Chatbots, supra note 14, at 11.

[59] See Carter, supra note 39, at 9 (“Every state has a consumer protection law that prohibits deceptive practices, and many prohibit unfair or unconscionable practices as well.”); Block, Inc., CFPB No. 2025-CFPB-0001, at 1 (Jan. 16, 2025) (finding that the “Respondent failed to provide effective customer service for Cash App in a manner that was unfair in violation of the Consumer Financial Protection Act (CFPA),” which contains an unfairness standard mirrored by many state UDAP laws); Press Release, Ga. Att’y Gen. Consumer Prot. Div., Carr Secures Settlement with Online Clothing Retailer Regarding Deceptive Advertising and Billing Practices (Nov. 7, 2025), https://consumer.georgia.gov/press-releases/2025-11-07/carr-secures-settlement-online-clothing-retailer-regarding-deceptive (describing a joint settlement finding that Target violated state consumer protection laws by “implementing and maintaining cancellation policies and practices that hindered consumers’ ability to cancel” its membership program).

[60] See 12 C.F.R. § 1024.35(e)(3) (requiring mortgage servicers to respond to a notice of error within 30 days of receipt, with a limited 15-day extension); 12 C.F.R. § 1005.11(c)(1)–(2) (requiring financial institutions to investigate and resolve an asserted electronic-transfer error within 10 business days, or within 45 days if the consumer’s account is provisionally credited in the interim); 15 U.S.C. § 1692g(b) (requiring debt collectors to cease all collection activity once a debt is disputed in writing, until the collector provides verification).

[61] Complaint ¶¶ 6–23, H&R Block Inc., No. 9427 (F.T.C. Feb. 23, 2024).

[62] Complaint ¶¶ 30–52, FTC v. 1661, Inc., No. 2:24-cv-10329 (C.D. Cal. 2024).

[63] Complaint ¶¶ 41–43, CFPB v. Navient Corporation, No. 3:17-cv-00101-RDM (M.D. Pa. 2017).

[64] CFPB Bans Navient from Federal Student Loan Servicing and Orders the Company to Pay $120 Million for Wide-Ranging Student Lending Failures, CFPB (Sep. 12, 2024), https://www.consumerfinance.gov/archive/newsroom/cfpb-bans-navient-from-federal-student-loan-servicing-and-orders-the-company-to-pay-120-million-for-wide-ranging-student-lending-failures/.

[65] CFPB Sues Comerica Bank for Systematically Failing Disabled and Older Americans, CFPB (Dec. 6, 2024), https://www.consumerfinance.gov/archive/newsroom/cfpb-sues-comerica-bank-for-systematically-failing-disabled-and-older-americans/.

[66] CFPB Orders Operator of Cash App to Pay $175 Million and Fix Its Failures on Fraud, CFPB (Jan. 16, 2025), https://www.consumerfinance.gov/archive/newsroom/cfpb-orders-operator-of-cash-app-to-pay-175-million-and-fix-its-failures-on-fraud/.

[67] Complaint No. 12120810, CFPB (Feb. 18, 2025), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/12120810.

[68] Complaint No. 5662117, CFPB (June 13, 2022), https://www.consumerfinance.gov/data-research/consumer-complaints/search/detail/5662117.

[69] See Carter, supra note 39, at 9.

[70] See 12 C.F.R. § 1024.35(e)(1)(i) (requiring mortgage servicers to respond to a notice of error by either correcting the identified error or “[c]onducting a reasonable investigation” and providing a substantive written determination); 12 C.F.R. § 1005.11(c) (requiring financial institutions to actually “investigate” an asserted electronic-transfer error and “determine whether an error occurred”).

[71] CFPB Takes Action Against Sutherland Global and NOVAD Management Consulting for Reverse Mortgage Servicing Failures, CFPB (June 18, 2024), https://www.consumerfinance.gov/archive/newsroom/cfpb-takes-action-against-sutherland-global-and-novad-management-consulting-for-reverse-mortgage-servicing-failures/.

[72] Complaint ¶ 40, South. Health Solutions, Inc., No. 232-3040 (F.T.C. July 14, 2025).

[73] DOT Penalizes Southwest Airlines $140 Million for 2022 Holiday Meltdown, U.S. Dep’t of Transp. (Dec. 18, 2023), https://www.transportation.gov/briefing-room/dot-penalizes-southwest-airlines-140-million-2022-holiday-meltdown.

[74] See Following Reports of “Bilt Card 2.0” Meltdown, Advocates Call on the Trump CFPB to Protect Borrowers, Warn Congress of Risks to Renters, Prot. Borrowers (Mar. 11, 2026), https://protectborrowers.org/following-reports-of-bilt-card-2-0-meltdown-cfpb-must-protect-borrowers-warn-congress/; Jason Mikula, Bilt Users Revolt Over Rocky Transition, Fintech Bus. Wkly. (Mar. 8, 2026), https://fintechbusinessweekly.substack.com/p/bilt-users-revolt-over-rocky-transition.

[75] Prot. Borrowers, supra note 74.

[76] To take another example: after a student loan borrower made a $500 payment to MOHELA that was never credited, the servicer issued a delinquency notice. Yet, MOHELA’s chatbot could not understand the issue—and hours on hold with customer service produced no resolution. Complaint No. 8313363, supra note 56.

[77] See Mishal Khan, Co-opting or Centering the Human Worker in the Digital Age? An Overview of Human-in-the-Loop Legislation in the United States, Just Tech (Aug. 5, 2026), https://just-tech.ssrc.org/articles/co-opting-or-centering-the-human-worker-in-the-digital-age-an-overview-of-human-in-the-loop-legislation-in-the-united-states/.

[78] Switching Systems Division, Rockwell Int’l, at 3 (1982).

[79] Samah Chowdhury, Behind the Bot: A History of Automated Customer Interaction Systems in Banking, Am. Bankers Ass’n (Oct. 25, 2024), https://bankingjournal.aba.com/2024/10/behind-the-bot-a-history-of-automated-customer-interaction-systems-in-banking/.

[80] From IVR to AI agents: 40 years of Promises that Call Centres Would Disappear, Smart Role (Nov. 24, 2025), https://www.smartrole.ai/blog/customer-support-automation-history.

[81] Oliver Teves, Call Center Jobs Drifting Overseas, CBS News (Dec. 9, 2003, at 12:31 ET), https://www.cbsnews.com/news/call-center-jobs-drifting-overseas/.

[82] Dion Jay Tality, 2025 Business Process Outsourcing Guide: Trends To Watch out For, Unity Commc’ns (May 25, 2025), https://unity-connect.com/our-resources/blog/bpo-industry/.

[83] Sean O’Brady, Virginia Doellgast & David Blatter, The High Costs of Outsourcing: Vendor Errors, Customer Mistreatment, and Well-Being in Call Centers, 63 Indus. Rels. 80, 82–85.

[84] Mass. Foreclosure Report, supra note 31, at 1 (“[U]nnecessary foreclosures had been occurring all too often because the system for servicing subprime mortgage loans was not designed to conduct large numbers of loan modifications or other work-outs for homeowners in distress.”); Larry Cordell, Karen Dynan, Andreas Lehnert, Nellie Liang & Eileen Mauskopf, Designing Loan Modifications to Address the Mortgage Crisis and the Making Home Affordable Program 3–4 (Fed. Res. Bd., Working Paper No. 2009-43, 2009) (“[T]he costs of loss mitigation are high and servicers have little financial incentive to invest heavily in the staff and technology to provide these services.”).

[85] In February 2026, the NYC Mayor shut down an AI chatbot that was functionally unstable and served inaccurate and illegal advice. Joe Wilkins, Mamdani Shuts Down NYC’s Disastrous AI Chatbot, Futurism (Feb. 2, 2026, at 16:09 ET), https://futurism.com/artificial-intelligence/ai-chatbot-mamdani. More action like this is needed.

[86] Joint Statement on Enforcement of Civil Rights, Fair Competition, Consumer Protection, and Equal Opportunity Laws in Automated Systems 2–3 (Apr. 4, 2024), https://www.justice.gov/archives/crt/media/1346821/dl (“Existing legal authorities apply to the use of automated systems and innovative new technologies just as they apply to other practices.”).

[87] Paul L. Singer, Abigail Stempson & Beth Bolen Chun, State Attorney General Civil Investigative Demands: Take Them Seriously, Kelley Drye (Mar. 15, 2023) (“[M]ost consumer protection laws give attorneys general broad authority to perform pre-litigation discovery through investigative subpoenas, often termed ‘civil investigative demands’ (CIDs). Many attorneys general can also require sworn statements and answers to interrogatories pursuant to this statutory authority.”).

[88] Cal. Fin. Code § 90011 (2025).

[89] Cal. Fin. Code § 90005 (2025).

[90] Stephanie T. Nguyen & Adam Teitelbaum, The Legal Risks Posed by AI Chatbots: Unfair and Deceptive Practices and Other Authorities 10–11, https://law-and-economy.law.columbia.edu/content/legal-risks-posed-ai-chatbots-unfair-and-deceptive-practices-and-other-authorities (proposing model document and information requests for regulators investigating a company’s use of AI customer service chatbots). Additionally, federal and state enforcement agencies can routinely maintain arrangements for sharing investigative, supervisory, and other nonpublic information with sister regulators—including federal agencies and state attorneys general and financial regulators. Those arrangements may allow an investigating agency to obtain relevant materials already produced to another regulator, potentially including records of chatbot customer-service interactions. See, e.g., CFPB, Policies & Procedures Manual 162–63 (2021) (describing the policy by which the CFPB may exchange confidential information with other agencies); Lina M. Khan, Rebecca Kelly Slaughter, Alvaro M. Bedoya, Melissa Holyoak & Andrew Ferguson, FTC, Working Together to Protect Consumers 11 (2024) (describing how the FTC collaborates with entities, including state attorneys general, by sharing information and expertise).

[91] A letter to leading AI companies about chatbot sycophancy by 42 State Attorneys General (AGs) outlines that “developers may be accountable for the outputs of their GenAI products.” Nat’l Ass’n Att’ys Gen., supra note 6, at 5. They also assert: “the onus falls on the company to continue to monitor for and remediate harmful outputs that endanger consumers.” Id. at 6. Public messaging like this is the right first step.

[92] See, e.g., Federal Trade Commission Announces Bipartisan Rule Banning Junk Ticket and Hotel Fees, FTC (Dec. 17, 2024), https://www.ftc.gov/news-event... (announcing bipartisan FTC “junk fees” rule that prohibits certain deceptive fee practices, prompting industry-wide change). The FTC “junk fees” rule inspired similar legislation and enforcement rules in New York. For example, New York City proposed a rule that would prohibit businesses from charging consumers hidden “junk fees” and require “all-in” pricing for products and services in the city. See N.Y.C. Dep’t of Consumer & Worker Prot., Notice of Public Hearing and Opportunity to Comment on Proposed Rules (June 16, 2026), https://www.nyc.gov/assets/dca/downloads/pdf/about/DCWP-NOH-Rules-Relating-to-Junk-Fees.pdf; Mayor Mamdani Announces Landmark “Click-To-Cancel” Consumer Protection Rules to Ban Subscription Traps and Junk Fees, N.Y.C. Off. of the Mayor (July 10, 2026), https://www.nyc.gov/mayors-office/news/2026/07/mayor-mamdani-announces-landmark--click-to-cancel--consumer-prot.

[93] See, e.g., Press Release, FTC, FTC Takes Action Against Amazon for Enrolling Consumers in Amazon Prime Without Consent and Sabotaging Their Attempts to Cancel (June 21, 2023), https://www.ftc.gov/news-events/news/press-releases/2023/06/ftc-takes-action-against-amazon-enrolling-consumers-amazon-prime-without-consent-sabotaging-their.

[94] The law’s protections are limited to businesses with more than $500 million in annual revenue and require only a good-faith effort—not an actual guarantee—to connect consumers with a human within 15 minutes. A.B. 1609, § 22627(b)(1), 2025–2026 Leg., Reg. Sess. (Cal. 2026), https://leginfo.legislature.ca.gov/faces/billVotesClient.xhtml?bill_id=202520260AB1609.

[95] Nicolas Spatola, AI Efficiency Can Undermine Accountability Even With Humans in the Loop, Tech Pol’y Press (May 5, 2026), https://www.techpolicy.press/ai-efficiency-can-undermine-accountability-even-with-humans-in-the-loop/.

[96] Seth Frotman, Hire 10,000 Corporate Cops, 80 Democracy J., Spring 2026, https://democracyjournal.org/magazine/80/hire-10000-corporate-cops/.

[97] See Hiring Technologists to Protect Consumers, CFPB, https://www.consumerfinance.gov/about-us/careers/cfpb-technologist/hiring-technologists-to-protect-consumers/; Memorandum from Lorelei Salas & Winston Berkman-Breen to Interested Parties 6 (Aug. 27, 2026), https://protectborrowers.org/wp-content/uploads/2026/08/Supervision-Policy-Blueprint.pdf; Government Agencies Act to Elevate and Build Tech and Digital Capacity, FTC (Mar. 26, 2024), https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/03/government-agencies-act-elevate-build-tech-digital-capacity.

[98] See, e.g., Erie Meyer, Complaints.gov: Building a Modern Infrastructure for Accountability (2025) (describing an exploratory concept envisioning infrastructure that connects people to the accountability systems we’ve already built and paid for).

[99] See Federal Agency Workforce Changes, U.S. Gov’t Accountability Off. (Feb. 24, 2026), https://www.gao.gov/products/gao-26-108719 (“In the first half of 2025, staffing declined at nearly all major agencies.”).

[100] Meyer, supra note 98, at 6.

[101] Some regulators have already published research questions. See, e.g., Solving the Traveling Salesman Problem? Not Quite, But Here Are More Research Questions from the Office of Technology, FTC (Dec. 4, 2024), https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/12/solving-traveling-salesman-problem-not-quite-here-are-more-research-questions-office-technology; P = NP? Not Exactly, But Here Are Some Research Questions From the Office of Technology., FTC (May 16, 2024), https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/05/p-np-not-exactly-here-are-some-research-questions-office-technology.

[102] Lina M. Khan, Chair, FTC, Remarks at the Harvard Center for Labor and a Just Economy 9 (Feb. 21, 2024), https://www.ftc.gov/system/files/ftc_gov/pdf/remarks-khan-harvard-center.pdf.