AI Under Trump: Deregulation, Preemption & the New Federal AI

Ai4

AI-assisted summary of AI Under Trump: Deregulation, Preemption & the New Federal AI — original source. Check the original for full context.

Source published: 2026-08-21

Detailed notes

## Background Published by Ai4 on YouTube on August 21, 2026, this panel features Trooper Sanders of the State AI Safety Roundtable and Jordan Crenshaw of the U.S. Chamber of Commerce. The discussion explores Washington's contentious AI policy landscape, focusing on federal deregulation, efforts to preempt state AI legislation, and the debate over who benefits and bears the costs when federal oversight recedes.

## TL;DR The panel debates Washington's push to preempt state-level AI regulations, contrasting concerns over small business compliance costs with the constitutional and systemic risks of leaving frontier AI unchecked. Jordan Crenshaw advocates for a four-step framework relying on technology-neutral enforcement and warns against costly state patchworks, while Trooper Sanders highlights how stripping state safety guardrails could trigger uninsurable commercial liabilities and overwhelm regulatory institutions. Despite differing starting points, both panelists agree that federal preemption via executive action is legally fragile and point toward 2027 for deliberate federal legislation, starting with foundational data privacy frameworks like the Secure Data Act.

## Atmosphere & style The panel maintains a measured, intellectually rigorous, and collegial atmosphere, eschewing partisan rhetoric in favor of deep legal and economic analysis. Despite representing potentially opposing interests—a civil society safety advocate and a senior business lobbyist—the speakers engage in a collaborative dialogue marked by mutual respect and surprising consensus. Listening to the exchange feels like sitting in on a high-stakes, behind-closed-doors Washington policy briefing that methodically strips away Silicon Valley hype to confront the grounded realities of constitutional federalism, administrative capacity, and insurance underwriting.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) Federal attempts to aggressively preempt state-level AI safety rules have faltered, sparking a rare bipartisan coalition of governors and state attorneys general defending state regulatory authority. - [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) Small businesses are adopting AI rapidly, with 58% utilizing the technology and 82% expanding their workforce, but 65% express deep concern over navigating a fragmented state-by-state patchwork. - [04:12](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=252s) State-level administrative rulemaking, such as California's expansion of the CCPA, has created multi-million-dollar economic burdens that disproportionately strain smaller enterprises. - [04:12](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=252s) Executive decrees attempting to preempt state AI regulations without comprehensive congressional statutory authorization face immediate constitutional challenges regarding state police and safety powers. - [06:46](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=406s) Without clear guardrails, commercial liability insurers may exclude frontier AI and autonomous agentic workflows, threatening to stall enterprise adoption across regulated sectors like healthcare and finance. - [08:50](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=530s) The U.S. Chamber advocates a four-step governance model: identifying concrete harms, applying existing laws, pinpointing specific statutory gaps, and deploying targeted, risk-based solutions. - [08:50](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=530s) Sweeping state omnibus mandates like Colorado's AI law risk creating substantial economic friction, projected by the Chamber as a $50 billion drag affecting 700,000 jobs. - [12:24](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=744s) Policy experts argue state lawmakers should prioritize enforcing technology-neutral consumer protection and anti-discrimination statutes rather than introducing duplicative, tech-specific legislation. - [12:24](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=744s) Enforcing AI safety requires expanding institutional capacity and resources across legacy agencies like the FDA and FTC before advanced recursive and autonomous systems trigger public safety crises. - [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) Data center developments face mounting local pushback over aesthetic concerns, grid strain, and rising utility rates, becoming the tangible focus of broader public anxiety surrounding AI. - [18:59](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1139s) Major regulatory frameworks require years of iterative negotiation to build durable policy, mirroring historical legislative efforts around mental health parity and prescription drug pricing. - [19:49](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1189s) Federal preemption is projected as more realistic by 2027, with comprehensive national data privacy legislation like the Secure Data Act serving as the foundational building block.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) Evaluating Washington's Approach to AI Regulation

The panel opens by framing the central conflict in contemporary artificial intelligence policy: the federal government's aggressive push to preempt state-level AI regulations. The moderator poses a foundational question to both panelists: is Washington’s current regulatory approach actually working? Trooper Sanders, President of the State AI Safety Roundtable, answers with an emphatic "spectacularly no." Sanders argues that federal preemption efforts have fallen completely flat, unintentionally forging a rare bipartisan coalition of governors and state attorneys general who refuse to be sidelined for the next decade. Drawing a parallel to civil aviation, Sanders asks the audience how many would board a completely unregulated aircraft, emphasizing that AI is "exponentially more complicated and dangerous than airplanes." He characterizes the Trump administration’s regulatory timeline as resembling a "student who didn't study for the bar exam until the night before," pointing to hastily assembled industry security agreements that should have been developed methodically over the past year.

Jordan Crenshaw, Senior Vice President at the U.S. Chamber of Commerce’s Technology Engagement Center, takes a more measured stance, calling the outcome "to be determined." However, Crenshaw highlights a fundamental misunderstanding among both Democrats and Republicans in Washington regarding the core definition and economic footprint of AI—a confusion mirrored in emerging state rules, such as recent debates in Florida. Crenshaw stresses that policy debates mistakenly focus almost entirely on Big Tech developers while ignoring Main Street. Small businesses are increasingly adopting AI to grow, but they face crippling compliance friction if subjected to a fragmented regulatory regime or ill-conceived federal mandates.

- **Small business AI adoption:** 58% of small businesses report using AI technologies. - **Workforce expansion:** Among small businesses utilizing AI, 82% are actively growing their workforce. - **Regulatory anxiety:** 65% of small business owners express concern over a confusing state-by-state patchwork of AI rules.

---

[04:12](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=252s) The Costs of State Patchworks and CCPA Implementation

When pressed on specific state laws harming American enterprise, Crenshaw points past Colorado to the California Consumer Privacy Act (CCPA) and its aggressive enforcement apparatus. He illustrates how the California Privacy Protection Agency took a single sentence of statutory text and expanded it into a ten-page, half-billion-dollar regulatory burden. Under federal standards, a $100 million economic footprint qualifies as a major rule; California’s administrative implementation quintupled that threshold in its interstate impact. Even CCPA’s original author, Alastair Mactaggart, questioned the legal validity of these promulgated rules during the voting process. Crenshaw warns that such rushed state rulemaking places an outsized financial burden on small enterprises that lack the resources of incumbent tech giants.

```mermaid flowchart TD accTitle: Economic Impact of State-Level AI Regulations subgraph CA["California: CCPA Expansion"] A["Single Statutory Sentence"] --> B["10-Page Administrative Rule"] B --> C["$500M Interstate Impact"] C --> D["$16,000 Annual Cost / Small Business"] end subgraph CO["Colorado: Proposed AI Act Impact"] E["State Compliance Mandates"] --> F["$50B National Economic Drag"] F --> G["700,000 Jobs Impacted"] end ```

Addressing the administration's proposed executive orders to preempt state AI rules, Sanders argues they would be "dead on arrival in court very fast." He explains that under constitutional federalism, core police powers, public health, and safety protections belong inherently to the states. While Congress has the constitutional authority to pass statutory preemption as part of comprehensive national legislation, an executive decree attempting to strip states of their regulatory authority before federal statutes are enacted represents an unconstitutional overreach that would prompt dozens of state attorneys general to file immediate legal challenges.

- **California compliance burden:** The California Privacy Protection Agency acknowledged compliance costs of approximately $16,000 annually per small business. - **Federal major rule threshold:** Federal guidelines classify any regulatory action exceeding $100 million as a major rule, a benchmark California's agency exceeded fivefold.

---

[06:46](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=406s) Insurance Insolvency and the Frontier AI Risk Landscape

Sanders addresses the tangible risks of stripping state authorities of oversight without robust federal guardrails. If preemption simply amounts to handing AI laboratories and frontier developers an unregulated "free reign," the economy faces severe systemic vulnerabilities—chief among them the uninsurability of advanced AI systems. Commercial liability insurers are already "blinking yellow" over recursive self-improvement and autonomous agentic workflows. Because underwriters cannot quantify, predict, or contain catastrophic liabilities associated with uncontrollable systems, they are threatening to exclude frontier AI from standard commercial liability policies.

Sanders warns that if the insurance industry steps back, the broader commercial ecosystem collapses: "AI may be innovative and disruptive. You know what's not? Insurance." Because frontier labs rely heavily on enterprise clients across highly regulated state sectors—such as healthcare, utilities, and financial services—enterprises facing tripled or quadrupled insurance premiums will simply abandon AI adoption. Without productive federalism that pairs national guidelines with state-level enforcement, market expansion will stall under the weight of uninsurable liabilities.

---

[08:50](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=530s) The U.S. Chamber's Four-Step Regulatory Framework

Rejecting the notion that the Chamber seeks blanket deregulation, Crenshaw outlines the policy framework established by a bipartisan commission co-chaired by former Representatives John Delaney and Mike Ferguson. The framework establishes a disciplined four-step test to govern artificial intelligence without stifling innovation or duplicating existing legal authorities across consumer protection and civil rights law.

```mermaid flowchart LR accTitle: U.S. Chamber Bipartisan 4-Step AI Governance Framework Step1["1. Identify Concrete Harms"] --> Step2["2. Apply Existing Laws"] Step2 --> Step3["3. Pinpoint Regulatory Gaps"] Step3 --> Step4["4. Deploy Targeted Risk Solutions"] ```

Crenshaw points out that all 50 states already enforce broadly applicable consumer protection and anti-discrimination statutes, alongside 25 states that have enacted comprehensive consumer data privacy laws. Where genuine legal gaps exist—such as non-consensual deepfakes addressed by the federal Take It Down Act—targeted, risk-based measures should be enacted. Conversely, sweeping omnibus mandates like Colorado's AI legislation create friction that disproportionately harms startups. Drawing on findings from the Chamber's *Empowering Small Business* report, Crenshaw notes that excessive compliance requirements force startups to burn capital on legal and compliance teams rather than product development, perversely cementing the dominance of market incumbents.

- **Economic exposure of state rules:** The Chamber projects that implementation of Colorado's AI law alone would create a $50 billion drag on the national economy and affect approximately 700,000 jobs. - **Small business sentiment:** Survey data reveals that one-third of small businesses expect state-level AI mandates to increase technology costs, while another third report they would be less likely to deploy AI tools altogether.

---

[12:24](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=744s) Technology-Neutral Enforcement vs. Institutional Capacity

When asked whether states should simply stand down while Washington deliberates, Crenshaw urges state lawmakers to focus on enforcing existing, technology-neutral statutes rather than drafting duplicative legislation. He observes that state legislators frequently introduce sweeping bills without conducting basic conflict checks against existing statutory authorities. Technology-neutral consumer protection frameworks naturally extend to future algorithmic breakthroughs—from generative models to quantum computing—preventing the need for legislative panic every time a new technology emerges.

Sanders agrees that existing legal frameworks provide a valuable baseline, but argues that statutes are useless without institutional enforcement capacity: "We need to have a consensus that government matters." He highlights the danger of under-resourced agencies trying to govern cutting-edge technologies. When agentic systems begin working alongside licensed healthcare professionals in clinical hospital environments, or when recursive AI tools—such as new ventures launched by Google co-founders—automate chemical and medical device discoveries at unprecedented speeds, legacy regulators like the FDA and FTC will be overwhelmed. Sanders warns that allowing public safety institutions to erode risks a catastrophic failure, triggering an aggressive populist backlash and rushed, reactionary emergency rulemaking.

---

[17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) Backlash Against Physical AI Infrastructure and Data Centers

The discussion shifts to the physical infrastructure powering artificial intelligence: the nationwide wave of community opposition to massive data center developments. Crenshaw states that the Chamber is not surprised by authentic local grievances regarding facility siting, aesthetic disruption, and electrical grid burdens driving up consumer utility rates. He observes that data centers have become the tangible, physical manifestations of broader public anxiety surrounding AI. However, Crenshaw also cautions that international adversaries and economic competitors are actively exploiting these local tensions to slow domestic infrastructure buildouts and undermine American competitiveness.

Reflecting on his prior experience in the White House Domestic Policy Council, Sanders concludes by comparing the current AI debate to historical regulatory transitions. He emphasizes that AI governance is "infinitely more complex" than past domestic policy challenges, expressing grave concern that policymakers are arbitrarily compressing their timeline to resolve foundational issues of safety, federalism, and liability.

[18:59](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1139s) The Inherent Deliberateness of Lawmaking: From Health Parity to AI

Reflecting on the pace of meaningful policy development, the discussion draws a historical parallel to the long, iterative fight for mental health parity in the 1990s—the principle that insurers must cover mental healthcare on equal terms with physical healthcare. That milestone required years of substantive negotiation, political maneuvering, and multiple revisions to get the policy mechanics right. A similar pattern was seen with recent congressional efforts to regulate complex prescription drug pricing regimes, which likewise demanded years of groundwork. Complex regulatory frameworks cannot be rushed, and AI safety is no exception: it cannot be effectively designed and rammed through in a matter of weeks without bringing key stakeholders, including the U.S. Chamber of Commerce and the broader business community, to the table.

This deliberate pace is not an accidental defect but a core feature of the American system. The nation's founders deliberately constructed the federal government to be slow, friction-filled, and frustrating to prevent hasty overreach and ensure thorough deliberation. AI policy must adapt to this constitutional architecture rather than bypass it, underscoring the urgent need for lawmakers and stakeholders to buckle down and do the necessary substantive work.

- **Historical Precedents**: Parallels drawn to 1990s mental health parity legislation and ongoing prescription drug pricing reforms, both taking multi-year cycles of drafting and revision. - **Institutional Design**: Acknowledgment that the legislative process was intentionally engineered by the founders to move slowly and withstand short-term impulses.

---

[19:49](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1189s) Outlook on Federal Preemption and Comprehensive Privacy Legislation

Turning to a final lightning-round question on whether a federal AI preemption statute will ever pass Congress, the panelists express genuine optimism. While acknowledging that passage is unlikely during the current congressional term, the outlook points toward 2027 as a realistic window for federal action.

However, the most immediate and foundational step must focus on comprehensive data privacy rather than direct AI mandates alone. Because artificial intelligence models depend entirely on the data used to train and run them, governing data inputs is the essential first building block. To establish a unified national standard, immediate attention must turn toward measures like the Secure Data Act, currently before the House of Representatives, to enact federal preemption and establish clear baseline rules for data governance. The session concludes with the moderator humorously noting that despite representing diverse regulatory and business viewpoints, the panelists found remarkable common ground across nearly every core policy question.

- **Preemption Timeline**: Anticipation that federal preemption legislation will not pass in the current Congress, but stands a strong chance by 2027. - **Priority Legislation**: The *Secure Data Act* in the House of Representatives highlighted as an essential preemptive foundation for governing the data underlying AI models.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) spectacularly no — Trooper Sanders

> [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) exponentially more complicated and dangerous than airplanes — Trooper Sanders

> [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) student who didn't study for the bar exam until the night before — Trooper Sanders

> [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) to be determined — Jordan Crenshaw

> [04:12](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=252s) dead on arrival in court very fast — Trooper Sanders

> [06:46](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=406s) blinking yellow — Trooper Sanders

> [06:46](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=406s) AI may be innovative and disruptive. You know what's not? Insurance. — Trooper Sanders

> [12:24](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=744s) We need to have a consensus that government matters. — Trooper Sanders

> [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) infinitely more complex — Trooper Sanders

## Mentioned Trooper Sanders, Jordan Crenshaw, State AI Safety Roundtable, U.S. Chamber of Commerce, U.S. Chamber Technology Engagement Center, Donald Trump, Joe Biden, Alastair Mactaggart, John Delaney, Mike Ferguson, California Privacy Protection Agency, California Consumer Privacy Act (CCPA), Colorado AI Act, Take It Down Act, Secure Data Act, Food and Drug Administration (FDA), Federal Trade Commission (FTC), White House Domestic Policy Council, Google, Empowering Small Business

## Video description Trooper Sanders - President, State AI Safety Roundtable Jordan Crenshaw - Senior Vice President, U.S. Chamber of Commerce

This is the most contested AI policy conversation in Washington right now. The administration's executive orders rescinding Biden's AI EO, the push to preempt state AI laws through the reconciliation bill, the dismantling of AI safety review processes, and the deregulatory posture toward Big Tech are generating active legislative conflict. This panel asks the defining question of 2026: who benefits when Washington steps back from AI oversight — and who bears the cost? Panelists represent the full spectrum: a former federal regulator who knows what is being dismantled, a civil-society leader on what deregulation means for communities, a competitive-district Republican navigating the Trump AI agenda, a think tank scholar on U.S. vs. global regulatory divergence, and a veteran Silicon Valley legislator delivering a 30-year perspective on what this moment actually means.

[Source](https://www.youtube.com/watch?v=kTRg8A6lcZU)

中文简报

特朗普治下的AI：放宽监管、联邦优先权与新联邦AI 专家组就华盛顿推动联邦法规优先于州级AI监管的举措展开辩论，对比了小企业合规成本压力与前沿AI放任不管所带来的宪法及系统性风险。Jordan Crenshaw主张推行依托技术中立执法的四步框架，并警告各州碎片化监管的高昂代价；Trooper Sanders则强调，剥夺州级安全防线可能会引发无法承保的商业责任，并使监管机构不堪重负。尽管出发点不同，两位专家一致认为通过行政手段推行联邦优先管辖在法律上相当脆弱，并预计审慎的联邦立法或将在2027年出台，且需从《安全数据法》（Secure Data Act）等基础数据隐私框架起步。联邦政府强推取代各州 AI 监管规则的举措遇阻，反倒促成了跨党派州长与总检察长捍卫地方监管权的联盟；目前 58% 的小企业已采用 AI，但其中 65% 对各州割裂的法规拼盘深感担忧。 (0:04) 加州 CCPA 扩增等州级规则带来数亿美元的合规重负，且每家小企业年均成本达 1.6 万美元；在缺乏国会法定授权下企图架空州警察权的总统行政令，将在法庭上迅速面临违宪挑战。 (4:12) 若剥夺州级监管而缺乏有效联邦护栏，商业责任险承保机构可能因无法量化风险而将前沿 AI 及自主代理系统排除在保单之外，进而威胁医疗与金融等受监管行业的应用落地。 (6:46) 美国商会提出由识别具体危害、适用现行法律、厘清法定空白及实施精准风控组成的四步治理框架，并警告科罗拉多州等泛化立法可能造成 500 亿美元的经济损失并冲击 70 万个就业岗位。 (8:50) 专家预计全面的联邦 AI 优先管辖法案更可能在 2027 年落地，而当下首要任务应是通过《安全数据法案》（Secure Data Act）等立法建立全国统一的数据隐私基准。 (19:49)

Reading summary

Policy debates over artificial intelligence center on a clash between federal efforts to preempt local rules and an expanding state regulatory patchwork. While State AI Safety Roundtable president Trooper Sanders argues that executive preemption is an unconstitutional overreach that risks market instability, U.S. Chamber of Commerce vice president Jordan Crenshaw contends that fragmented state mandates impose severe compliance burdens on small businesses. Constitutional Hurdles to Executive Preemption (4:12) Sanders contends that the Trump administration's strategy of using executive orders to preempt state AI regulations will fail rapidly in court. Under American constitutional federalism, core police powers, public health, and safety protections belong inherently to the states. While Congress holds the authority to enact statutory preemption through comprehensive national legislation, Sanders asserts that unilateral executive decrees attempting to strip states of regulatory power represent an overreach. This push has provoked an unusual bipartisan resistance among state attorneys general and governors who refuse to surrender their oversight roles. Sanders characterizes the administration's regulatory approach as rushed and reactive, comparing it to cramming for an exam rather than systematically building durable standards alongside state authorities. Compliance Burdens and Fragmented State Rules (0:04) Crenshaw warns that regulatory discourse disproportionately focuses on Big Tech incumbents while ignoring small enterprises. According to Chamber data, 58 percent of small businesses use AI technologies, and 82 percent of those adopters are growing their workforces. However, 65 percent of small business owners report anxiety over conflicting state-by-state mandates, which can divert limited capital toward legal compliance rather than innovation. To illustrate the cost of state-level overreach, Crenshaw cites the California Privacy Protection Agency's expansion of a single sentence in the California Consumer Privacy Act into a ten-page rule with a half-billion-dollar interstate impact—exceeding the federal $100 million major-rule threshold fivefold and costing small businesses roughly $16,000 annually. Additionally, the Chamber projects that Colorado's AI legislation could impose a $50 billion drag on the national economy and impact 700,000 jobs, with one-third of surveyed small firms indicating state rules would deter them from deploying AI. Liability Underwriting and Uninsurable Frontier AI (6:46) Sanders cautions that deregulating frontier AI without federal guardrails exposes the commercial economy to severe structural risks, specifically insurance market insolvency. Commercial underwriters are growing increasingly cautious about recursive self-improvement and autonomous agentic workflows because catastrophic risks from uncontained systems cannot currently be quantified or predicted. If insurers exclude advanced AI from commercial liability policies, enterprise adoption could collapse across regulated sectors such as healthcare, financial services, and public utilities. Sanders emphasizes that enterprises facing soaring insurance premiums will halt deployment, arguing that sustainable market expansion depends on pairing federal baselines with state enforcement to keep frontier systems insurable. Technology-Neutral Enforcement and Privacy Baselines (8:50) To prevent legislative overreach, Crenshaw advocates a four-step framework developed by a bipartisan Chamber commission: identify concrete harms, enforce existing laws, pinpoint genuine regulatory gaps, and deploy targeted remedies. He argues that existing, technology-neutral consumer protection and anti-discrimination statutes already apply across algorithmic systems, avoiding the need for duplicative state mandates. Sanders agrees that existing statutes provide a baseline, but warns that legacy agencies like the FDA and FTC lack the institutional capacity and funding to oversee autonomous agentic systems in critical areas like clinical medicine or chemical discovery. Looking ahead, both panelists anticipate that statutory federal preemption will not pass during the current Congress, pointing to 2027 as a realistic window and identifying national data privacy legislation, such as the Secure Data Act, as the essential prerequisite for regulating AI systems. The panelists conclude that lasting AI governance requires deliberate, multi-year legislative work rather than hasty executive preemption or fragmented state mandates. However, achieving stability depends on whether Congress can enact baseline federal privacy standards by 2027 while adequately resourcing public agencies to supervise increasingly complex, agentic technologies.

精炼阅读摘要

州AI安全圆桌会议主席Trooper Sanders与美国商会高级副总裁Jordan Crenshaw围绕联邦优先统辖权（Preemption）与州级人工智能规制展开辩论。双方在行政命令统辖的宪法合法性与规制深度上存在明显分歧，但一致认为零散的州级法案正严重阻碍小微企业创新，而建立全国统一框架的前提是依循审慎的立法程序并优先出台联邦数据隐私法案。联邦行政统辖的宪制困境与博弈 (0:04) Sanders指出，特朗普政府试图通过行政命令强行剥夺各州AI监管权的策略在司法层面将“迅速夭折”。依据美国宪法联邦制原则，核心治安权、公共卫生与安全防护职能属于各州管辖。在国会尚未通过综合性联邦法案之前，单纯依靠白宫行政命令剥夺州级权力属于违宪越权，已经促使两党州长与州总检察长组成罕见的联盟准备提起诉讼。Sanders将政府匆忙拼凑行业自律协议的做法比作“考前突击备考”，并强调AI的技术复杂度与危险性远超民用航空，绝不能以去监管为由放任其裸奔。 Crenshaw则对联邦统辖前景持相对审慎的“有待观察”态度，但他同样批评华盛顿两党对AI的核心技术定义及其真实经济体量缺乏基本理解，这种误区也普遍蔓延在佛罗里达等州的立法辩论中。Crenshaw认为，联邦决策层不能将目光局限于少数科技巨头，若监管政策持续混淆各方权责且漠视实体经济，将导致制度预期进一步恶化。州级规制碎片化对小微企业的合规冲击 (4:12) Crenshaw援引美国商会数据指出，全美已有58%的小微企业采用AI技术，且其中82%的企业正借此扩招员工；然而，65%的小企业主对各州碎片化、拼图式的法规深感焦虑。以加利福尼亚州《消费者隐私法案》（CCPA）为例，加州隐私保护局将法条中的单一句子扩展为长达10页的繁复细则，导致每家小企业每年面临约16,000美元的直接合规成本。按照联邦标准，经济影响超1亿美元即构成重大规则，加州该行政细则的跨州影响达5亿美元，超标五倍，甚至CCPA法案原起草人Alastair Mactaggart也在表决中对其法律有效性提出了质疑。此外，针对科罗拉多州提议的《人工智能法案》，美国商会测算其全面实施将对全美经济造成约500亿美元的拖累，并影响约700,000个工作岗位。调查显示，三分之一的小微企业预计州级指令将直接推高技术采购成本，另有三分之一明确表示可能因此暂缓或放弃部署AI工具。高昂的合规成本迫使初创企业将宝贵资本用于雇佣法务团队而非技术研发，客观上强化了老牌科技垄断者的竞争壁垒。系统性风险、商业保险失效与执法机构短板 (6:46) Sanders重点警示了单方面剥夺各州监督权可能引发的系统性金融脆弱性，尤其是前沿AI系统的“不可保性”。面对递归自我改进与自主智能体（Agentic workflows），商业责任保险承保人已对不可测风险亮起“黄灯”。由于精算师无法对失控系统造成的灾难性潜在责任进行量化和兜底，保险机构正威胁将前沿AI彻底排除在商业责任险范围之外。由于前沿模型实验室高度依赖医疗、公共事业和金融等受强监管的州级企业客户，一旦企业面临保费飙升数倍甚至断保，必将终止部署AI，导致整个商业化扩张陷入停滞。与此同时，Sanders强调技术中立的既有法律若缺乏充足的执行机构资源，无异于废纸一张。随着具备自主行动能力的AI进入医院临床辅助医疗，或递归算法以超常速度自动推进化学与医疗器械研发，美国食品药品监督管理局（FDA）和联邦贸易委员会（FTC）等传统监管机构的人员与技术储备将难以招架。倘若公共安全守护机构被持续削弱，一旦发生重大系统性事故，极易诱发强烈的民粹主义反弹与仓促过激的被动应急立法。四步法框架、物理基建阻力与立法展望 (8:50) 针对监管路径，Crenshaw否定了全面去监管的诉求，介绍了由前众议员John Delaney和Mike Ferguson共同牵头的两党“四步规制框架”：首先识别具体损害，其次运用50个州既有的反歧视与消费者权益法律以及25个州的通用隐私法，第三精准找出监管空白（例如针对非自愿深度伪造的联邦《Take It Down Act》），最后再实施分级定向风险治理。此外，AI对电力和物理土地的消耗正引发全美范围内对数据中心建设的抵制潮，居民对电费上涨和审美破坏的真实不满正在显现，Crenshaw提醒国际竞争对手亦在借机激化本土矛盾以拖延美国关键基建建设。在规制出台时间线上，讨论回顾了1990年代美国心理健康平价法案以及近年处方药价格改革历经数年的漫长博弈，指出美国宪政体制本身即被设计为审慎、缓慢且充满阻力，AI规则的制定无法在数周内速成。两位专家一致预计，现任国会无法完成联邦AI优先统辖法案的立法，2027年才是更切实的窗口期。当前最紧迫的突破口在于数据治理，应优先推进众议院审议的《安全数据法案》（Secure Data Act）等全面数据隐私法，为全国AI规则奠定底座。本场对话反映出，尽管安全阵营与工商业界在行政命令统辖的效力边界和执法机构资源配比上针锋相对，但双方均反对地方规制碎片化带来的成本攀升，并认同AI立法的审慎复杂性。尚未解决的核心问题在于：在国会预期于2027年采取实质行动前的监管真空期内，联邦与各州将如何在司法诉讼的拉锯中维持动态平衡，以及商业保险机构与传统监管机构能否在前沿智能体引发系统性事故前建立有效的缓冲防线。

Detailed notes — zh

## Background 本内容由 Ai4 频道于 2026 年 8 月 21 日发布在 YouTube 平台上。页面简介指出，该对谈围绕特朗普政府时期华盛顿最具争议的 AI 政策展开，重点探讨废除拜登 AI 行政令、通过预算法案推行联邦优先权以废止各州立法、削减 AI 安全审查以及对科技巨头采取去监管化立场等议题，并邀请了代表州级安全圆桌会与美国商会的政策专家，共同讨论在联邦监管后退的背景下究竟谁获益、谁承担代价。 ## TL;DR 本期节目深入探讨了华盛顿联邦 AI 优先权（Preemption）策略与各州分散立法之间的激烈博弈。美国商会代表 Jordan Crenshaw 指出，加州和科罗拉多等州各自为政的“拼贴式”繁琐监管严重挤压了中小企业的生存空间与技术采用率，因而极力呼吁出台统一的技术中立型国家级联邦标准。州级 AI 安全圆桌会主席 Trooper Sanders 则直言联邦单方面强推优先权在宪法与商业逻辑上均行不通，特别警示前沿递归自我改进 AI 若缺乏健全监管，将引发商业责任保险市场的承保危机与系统性风险；双方最终在“善用现行法律”与“强化监管机构专业执行能力”的必要性上达成了理性共识。 ## Atmosphere & style 整场对谈气氛紧凑、专业且交锋激烈，既充满华盛顿顶层政策博弈的针锋相对，又保持了务实理性的跨界对话风度。两位嘉宾站在完全不同的利益与制度视角——一方代表商业实体的合规与运营痛点，另一方代表州级安全防线与宏观系统性风险，言辞犀利却屡有幽默互动与罕见的理念共鸣。聆听过程中能直观感受到 AI 治理在联邦宪政制衡、前沿技术迭代与现实经济底座之间剧烈碰撞的紧迫感与复杂性。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) 联邦政府强推优先权（Preemption）以废止各州 AI 立法的做法遭遇强烈阻力，反向促成了由跨党派州长与总检察长组成的联合抗议阵线。 - [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) 华盛顿政界对 AI 基础定义存在认知偏差，缺乏长期治理规划而仓促拼凑的安全协议无法提供稳定的监管预期。 - [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) 调研显示 58% 的受访小企业已应用 AI 且多数在扩招员工，但高达 65% 的企业极度担忧各州碎片化法律拼凑带来的高昂合规成本。 - [04:12](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=252s) 加州隐私保护局（CPPA）将单句法案扩写为成本高达 5 亿美元的规则，每年给小企业增加约 1.6 万美元负担，成为地方监管过度膨胀的典型例证。 - [05:45](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=345s) 若在缺乏国会正式立法的前提下仅凭行政命令剥夺各州治安警察权，将引发全美 50 州联名违宪诉讼，在司法层面面临流产风险。 - [05:45](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=345s) 商业责任保险是实体经济底座，面对代理型 AI（Agentic AI）和递归自我改进的不可控风险，险企可能拒保或暴涨保费，从而切断前沿 AI 的商业化土壤。 - [08:50](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=530s) 美国商会提出四步监管法则：锚定真实伤害、穷尽现行法律、识别法律真空、制定分级风险方案，主张采取技术中立而非盲目全面去监管化。 - [08:50](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=530s) 据商会测算，仅科罗拉多州一地过激的 AI 立法就可能造成全美 500 亿美元经济损失并波及 70 万就业岗位，导致三分之一的小企业直接放弃采用技术。 - [11:31](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=691s) 割裂的各州监管规则客观上偏袒了有能力雇佣法务团队的科技巨头，严重削弱了初创公司与中小企业利用 AI 降本增效的竞争力。 - [13:58](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=838s) FTC 与 FDA 等核心监管机构在 AI 时代严重缺乏专业人才与“执行肌肉”，面对 AI 自动化科研每周涌现的新产品，传统审批机制面临崩溃风险。 - [13:58](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=838s) 监管缺位一旦造成实际公众伤害，将迅速激化民粹主义政治反弹，迫使政客在恐慌中出台质量低劣的应急性惩罚法规。 - [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 全美蔓延的反数据中心浪潮，是公众对 AI 占用土地、推高居民电费等现实隐患的具象化映射，也掺杂了地缘政治竞争对手借题发挥的因素。 - [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 重大体制改革历来耗时漫长，AI 治理复杂度呈几何级数上升但时间窗口极窄，试图在数周内仓促过关并不符合立法客观规律。 - [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 嘉宾预测联邦优先权法案虽难以在本届国会落地，但有望在 2027 年实现突破；目前最可行且紧迫的制度基石是全面的联邦数据隐私立法。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) 华盛顿的监管策略是否奏效？两方阵营的首轮交锋小组成员聚焦于特朗普政府治下的 AI 政策、去监管化（deregulation）以及联邦优先权（state preemption，即联邦法律凌驾并废止各州立法）的争议。主持人提出核心议题：联邦政府正发起阻断各州独立立法的行动，这直接决定了未来全美 AI 的治理路径。面对“华盛顿目前的监管策略到底管用吗？”这一开场问题，两位嘉宾给出了截然相反的定性。 ```mermaid flowchart TD accTitle: 华盛顿AI监管路径争议 subgraph Federal["华盛顿联邦层面"] A["联邦优先权预占策略 (Preemption)"] B["仓促出台的行业安全协议"] end subgraph States["州层面与微观实体"] C["跨党派州长与总检察长联盟反对"] D["58% 已应用 AI 的中小企业"] E["保险市场：面临承保不能的系统性风险"] end A -->|"剥夺10年立法权引发反弹"| C A -->|"若无联邦底线直接放权大模型"| E B -->|"未能解决概念定义与现有法律重合"| D ``` 州级 AI 安全圆桌会议（State AI Safety Roundtable）总裁 Trooper Sanders 毫不留情地直言：“简短的回答是，极其彻底地不管用（spectacularly no）。”他认为华盛顿推行的联邦优先权举措迄今为止完全碰壁，甚至反向促成了一个由跨党派州长、州总检察长等地方官员组成的联合阵线。各州强硬表态：“等一下，你们不能勒令我们在未来十年内对 AI 彻底无所作为。”此外，Sanders 打了个生动的比方：大多数参会者都是搭飞机来的，但谁敢坐一架毫无安全监管的飞机？AI 的复杂与危险程度比民航飞机高出指数级，必须建立确定的监管规则。然而，华盛顿目前的表现就像是一个“在司法考试前夜才临时抱佛脚的学生”——政府突然匆忙抛出一份与科技行业签署的“新 AI 安全协议”，尽管协议本身属于监管行为，但它本应在一年前就稳步推进，而不是在短短几周内草草拼凑上线。美国商会（U.S. Chamber of Commerce）科技参与中心高级副总裁 Jordan Crenshaw 则认为效果“尚待观察（to be determined）”。他指出华盛顿两党政客目前对 AI 存在深层的认知偏差，甚至在最基础的概念界定上都一头雾水（例如当天佛罗里达州刚刚提出的一项新规，连 AI 的定义都尚未厘清）。更严重的是，国会完全低估了法规对小型实体的附带伤害——监管不只是针对科技巨头开发者，全美使用 AI 的中小企业正在迅速扩大，却被迫暴露在混乱规则的阴影之下。此外，国会普遍忽视了 AI 在许多领域早已受到现行法律的管辖，并非处于法律真空状态。 - 58% 的受访小型企业目前已在业务中应用 AI 技术。 - 在使用 AI 的小型企业中，有高达 82% 正在扩大其员工雇佣规模。 - 65% 的小型企业高度担忧各州立法各自为政形成的“碎片化拼凑”（patchwork of rules）所带来的合规风险。 --- ### [04:12](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=252s) 加州隐私新规的失控与州立法的附带损害针对美国商会一直呼吁建立统一的“联邦国家级单一法规”的立场，主持人询问是否存在某一特定州法正在实质伤害美国企业。Jordan Crenshaw 坦言，此前他们最担忧的是科罗拉多州的 AI 法案，但就目前仍在生效的法律而言，最令人警惕的是《加州消费者隐私法》（CCPA）及其监管机构的执行方式。 Crenshaw 举了一个监管过度膨胀的典型例子：加利福尼亚州隐私保护局（CPPA）将 CCPA 文本中仅有的一句话，自行扩写变成了一部长达 10 页、合规成本高达 5 亿美元的重磅规则。在联邦层面上，经济影响达 1 亿美元的规定就会被正式定性为“重大规则”（Major Rule），而加州隐私局出台的这一项规则对州际商业的影响力直接翻了整整五倍。加州隐私局自身甚至公开承认，小型企业每年仅为遵守这一揽子庞杂的隐私与 AI 新规，就需额外承担 16,000 美元的支出。更荒谬的是，就连 CCPA 法案的原作者 Alistair MacTaggart 本人在表决时都公开发出质疑，怀疑这些衍生规则是否具备合法性，整场投票沦为一种“先投出来看看会发生什么”的草率实验。Crenshaw 强调，各州在不理解对实体经济和小型业主破坏力的情况下竞相仓促立规，这是极度危险的趋势。 --- ### [05:45](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=345s) 宪政危机与商业保险市场的断裂风险针对 Trooper Sanders 曾公开撰文断言“特朗普关于联邦优先权的行政令在法庭上将迅速胎死腹中（dead on arrival）”，Sanders 详细阐释了法理逻辑与现实商业后果。从宪政结构来看，直接剥夺各州一切监管权限公然违背了建国先贤的初衷与宪法原则——宪法明确将维护公共卫生、人身安全以及“治安警察权”（policing power）赋予了各州。如果在没有任何豁免保护条款的情况下强推优先权，将引发全美 50 个州联名上诉司法系统。因为按照正常的联邦制运转机制，优先权必须建立在国会已正式通过具体立法的前提之下，而不是通过行政命令直接叫停地方权力。 ```mermaid flowchart LR accTitle: 前沿AI与保险系统性崩塌链条 subgraph FrontierAI["前沿 AI 风险特征"] F1["代理型/智能体 AI (Agentic AI)"] F2["递归式自我改进 (Recursive Self-Improvement)"] end subgraph Insurer["商业责任保险业"] I1["无法理解风险"] I2["无法量化并遏制敞口"] I3["拒绝承保或保费暴涨 3-5 倍"] end subgraph Economy["实体经济连锁反应"] E1["中小企业失去责任保障，停止部署"] E2["前沿模型实验室收入枯竭，商业逻辑破裂"] end FrontierAI --> I1 & I2 I1 & I2 --> I3 I3 --> E1 --> E2 ``` 更核心的潜在灾难在于经济系统的底座——商业保险机制。Sanders 向全场抛出一个尖锐的反问：“如果你们的责任保险公司明天停止向你们提供承保，在座有几家企业还能维持运转？”他透露，商业保险巨头们目前正对前沿 AI（Frontier AI）集体亮起黄灯警报。面对具备自主决策能力的代理型 AI（Agentic AI）以及能够自我迭代优化的递归式自我改进（recursive self-improvement），承保人发现这些技术风险无法度量、无法限制，因此未来可能根本无法理赔。 Sanders 强调：“没有保险，现代经济体制就根本不复存在。AI 固然充满创新与颠覆性，但保险业可不是供你玩颠覆的行当。”一旦联邦层面简单粗暴地废除各州的审慎规则，转而放任模型实验室和行业野蛮生长，带来的绝非政治博弈，而是大批大中小企业因无法获得保单或面临暴涨三至五倍的保费，从而不得不全面停用 AI 产品。除了军工国防之外，AI 实验室的大多数潜在客户在州一级都处于高度监管之下；如果客户失去了承保能力，前沿 AI 实验室将彻底失去商业回血的市场土壤。 --- ### [08:50](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=530s) 美国商会的四步监管框架与过激立法的经济代价面对主持人关于“商会是否只是一味为巨头主张全面去监管化”的质疑，Jordan Crenshaw 澄清，商会对待 AI 的态度绝非盲目的“去监管化（deregulatory）”，而是提倡建立在现有法律土壤上的理性框架。他回顾了数年前由前民主党国会议员 John Delaney 与前共和党国会议员 Mike Ferguson 牵头组成的跨党派委员会所制定的四步应对法则： 1. **锚定真实伤害**：首先明确界定技术在实际应用中试图解决或引发的具体危害。 2. **穷尽现行法律**：审视现行法典是否已能涵盖该场景。事实上，全美每一个州均有普遍适用的消费者保护法、民权法和反歧视法；同时已有 25 个州通过了全面的数据隐私法，现有法律体系本就具备管辖力。 3. **识别法律真空**：精准找出法律规范确实落后于技术发展的真空地带。 4. **制定分级风险方案**：针对真正的制度空白，基于风险等级建立精准监管措施（例如已正式立法生效、将深度伪造报复性图像定罪的《Take It Down Act》）。 Crenshaw 指出，小企业生存最艰难的阻力就在于疲于应付各州割裂的法律补丁。美国商会在其发布的年度《赋能小企业报告》（Empowering Small Business Report）中曾将科罗拉多州的 AI 法案下发给企业主进行调研，企业给出了极其悲观的预期反馈： - 约三分之一（1/3）的小型企业明确表示，若各州该类法案生效，其采购与部署 AI 的成本将显著增加。 - 另有约三分之一（1/3）的小企业主坦言，在严苛复杂的监管重压下，他们将直接放弃采用 AI 技术，导致技术采用率暴跌。 - 据商会模型测算，仅科罗拉多州一地的激进 AI 立法如果全面付诸实施，将对全美经济造成高达 **500 亿美元** 的潜在负面冲击，并波及约 **700,000 个** 就业岗位。 ### [11:31](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=691s) 州级碎片化监管对小企业的伤害与全国性框架诉求 Jordan Crenshaw（美国商会高级副总裁）深入剖析了各州单独出台 AI 立法对商业生态造成的负面冲击。他以科罗拉多州的 AI 法案为例指出，仅这一部法律的合规要求就足以让小企业不堪重负。初创公司和小企业引入 AI 的初衷是为了削减开支、提高效率、快速起步并在市场中参与竞争。然而，当这些企业为了应对复杂的法律条文而不得不聘请庞大的律师团队和合规分析师时，使用 AI 带来的降本增效优势便荡然无存。Crenshaw 强调，碎片化的“拼贴式”（patchwork）监管客观上只对资源雄厚的大公司和既得利益行业巨头有利，因为只有它们才有能力承担高昂的合规成本。正因如此，美国商会坚定主张在数据隐私和人工智能两大领域建立统一的国家级标准与联邦框架，实行全国通用的单一规则体系。与此同同时，Crenshaw 澄清商会并非主张完全剥夺州政府的职权。各州拥有大量现成且行之有效的“通用法律”（laws of general applicability），这些法律完全可以用于应对当下的 AI 挑战。主持人随即抛出疑问：既然多州州长辩称各州立法是因为联邦政府迟迟不作为，那么商会是否是在要求各州在未来三到四年内袖手旁观、干等联邦行动？Crenshaw 回应称，州议员应当采取商会提出的务实框架：在仓促推进全面 AI 法案之前，先深入自查现有州法中是否有适用的条款来治理 AI 带来的实际危害。他透露，商会曾多次与起草法案的州议员沟通并询问：“你们做过与现行法律的冲突审查吗？查过这些问题是否能直接用现有州法解决吗？”结果往往是否定的。Crenshaw 主张采取“技术中立”（technology neutral）的立法原则，利用现有反欺诈、反歧视等成熟法律去惩处技术引发的消费者损害。这种做法能避免立法机构陷入每出现一种新技术（例如未来的量子计算）就不得不重新立法的恶性循环。 ### [13:58](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=838s) 监管机构的“执行肌肉”匮乏与制度风险 ```mermaid flowchart TD accTitle: AI 加速发展下的监管机构压力与社会反弹循环 subgraph S1["技术发展与审批压力"] A["前沿 AI 加速科研与产品发现<br/>（如递归自我改进 AI）"] --> B["每周涌现大量新化学品/医疗设备"] B --> C["监管机构审批过载与执法能力崩溃<br/>（FTC/FDA 长期缺乏资源与专业技能）"] end subgraph S2["社会与政治连锁反应"] C --> D["公众安全受到实质威胁与危害事件发生"] D --> E["引发民粹主义政治反弹（Populist Backlash）"] E --> F["政客仓促通过应急立法（Bad Policy）"] end ``` 当主持人询问现有的联邦贸易委员会（FTC）、食品药品监督管理局（FDA）等机构是否足以应对 AI 风险时，Trooper Sanders（州级 AI 安全圆桌会主席）给出了坚决的否定回答。虽然他与 Crenshaw 在“现有法律体系具备巨大治理潜力”这一层面上达成了罕见的共识，引得现场一阵笑声，但他尖锐地指出问题的核心在于：政府机构是否拥有让这些法律生效的“执行肌肉”（muscle）与专业能力。Sanders 强调，全美必须就“政府监管至关重要”（government matters）达成社会共识。即便各州拥有健全的法律，但如果执法机构缺乏懂技术的专业人员，法律就会形同虚设。例如，几十年前制定的护士执业许可法规，根本无法应对如今智能体 AI（Agentic AI）与医生、护士并肩协作的临床环境。一旦监管缺位导致安全事故，必然会“毒化 AI 的政治生态”，进而激起民粹主义反弹，迫使政客们在恐慌中草率出台劣质的应急政策。 Sanders 特别点名了两个关键监管领域： - **联邦贸易委员会（FTC）：** 作为一个本应极其强大且具有影响力的机构，FTC 在过去数十年中被“蓄意削弱”，这种虚弱状态在 AI 时代极其危险。 - **医疗与科学审批（FDA 等）：** Google 联合创始人近期离职并创立了一家专注于“递归自我改进 AI”（recursive self-improvement AI）以实现科学发现自动化的新公司。这意味着由 AI 自动推导的新化学品、新医疗设备将以每周为单位海量涌现。现有监管人员连应对人类研发的产品都捉襟见肘，若不强化机构能力，整个审查系统将面临崩溃，进而将公众置于险境。 Sanders 总结道，美国确实需要聪明的联邦法律、自愿性行业规范以及国际标准，但各方应当放平心态（"using our inside voices"）、扎实做好功课，由国会、白宫与各州州长展开持续的理性沟通与协作。 ### [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 数据中心社会抵触潮、历史监管立法周期与联邦优先权展望在讨论全美各地迅速蔓延的“反数据中心”（anti-data center）浪潮时，Jordan Crenshaw 表示商会对此并不感到意外。居民对后院新建大型设施、电费费率（ratepayers）上涨等民生问题抱有真实且合理的担忧。数据中心实际上是公众对人工智能潜在风险在物理现实层面的直接具象化映射（"manifestation physically"）。不过，Crenshaw 也指出，部分外部竞争对手正在借题发挥，试图利用这些民意阻力来遏制美国在 AI 领域的整体竞争力；这种抗议浪潮蔓延的速度之快确实超出了预期。随后，主持人询问曾担任白宫国内政策顾问的 Trooper Sanders，本轮 AI 监管战役与以往的历史立法相比有何不同。Sanders 直言，AI 治理的复杂度呈几何级数上升（"infinitely more complex"），但各方留给自己的时间窗口却极其狭窄。他以自身 1990 年代参与推动的“心理健康平权”（Mental Health Parity，即保险公司须对心理与生理健康提供同等报销）法案为例，说明重大政策往往需要经历数年乃至数十年的实质修订与政治博弈；包括国会当天正在审议的一项针对处方药定价机制的法案同样耗时多年。美国的建国先贤们将政府体制设计得“缓慢而令人沮丧”（slow and frustrating），AI 监管不可能在几周内仓促强推过关，必须广泛吸纳企业及各方意见。在最后的总结环节，主持人向两位嘉宾提出了关于“联邦预先推翻/优先权法案（Federal Preemption Bill）能否在国会通过”的终极问题： - **Trooper Sanders 的判断：** 坚信联邦优先权法案必然会通过（"Oh, absolutely"），虽然在本届国会难以落地，但有望在 2027 年见分晓。 - **Jordan Crenshaw 的补充：** 认为目前最可行且最紧迫的突破口是全面的数据隐私立法。因为监管流入 AI 系统的训练数据是治理的基石，众议院目前正在审议的《数据安全法案》（Secure Data Act）等具备优先权的立法应当率先通过，以奠定第一块制度基石。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) 简短的回答是，极其彻底地不管用（spectacularly no）。 — Trooper Sanders > [00:04](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=4s) 大多数参会者都是搭飞机来的，但谁敢坐一架毫无安全监管的飞机？AI 的复杂与危险程度比民航飞机高出指数级，必须建立确定的监管规则。 — Trooper Sanders > [05:45](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=345s) 如果你们的责任保险公司明天停止向你们提供承保，在座有几家企业还能维持运转？ — Trooper Sanders > [05:45](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=345s) 没有保险，现代经济体制就根本不复存在。AI 固然充满创新与颠覆性，但保险业可不是供你玩颠覆的行当。 — Trooper Sanders > [11:31](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=691s) 碎片化的“拼贴式”（patchwork）监管客观上只对资源雄厚的大公司和既得利益行业巨头有利，因为只有它们才有能力承担高昂的合规成本。 — Jordan Crenshaw > [11:31](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=691s) 你们做过与现行法律的冲突审查吗？查过这些问题是否能直接用现有州法解决吗？ — Jordan Crenshaw > [13:58](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=838s) 美国确实需要聪明的联邦法律、自愿性行业规范以及国际标准，但各方应当放平心态（"using our inside voices"）、扎实做好功课，由国会、白宫与各州州长展开持续的理性沟通与协作。 — Trooper Sanders > [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 数据中心实际上是公众对人工智能潜在风险在物理现实层面的直接具象化映射（"manifestation physically"）。 — Jordan Crenshaw > [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 美国的建国先贤们将政府体制设计得“缓慢而令人沮丧”（slow and frustrating），AI 监管不可能在几周内仓促强推过关，必须广泛吸纳企业及各方意见。 — Trooper Sanders > [17:19](https://www.youtube.com/watch?v=kTRg8A6lcZU&t=1039s) 监管流入 AI 系统的训练数据是治理的基石，众议院目前正在审议的《数据安全法案》（Secure Data Act）等具备优先权的立法应当率先通过，以奠定第一块制度基石。 — Jordan Crenshaw ## Mentioned Trooper Sanders, Jordan Crenshaw, State AI Safety Roundtable, U.S. Chamber of Commerce, CPPA, CCPA, Alistair MacTaggart, John Delaney, Mike Ferguson, Take It Down Act, FTC, FDA, Google, Secure Data Act ## Video description Trooper Sanders - President, State AI Safety Roundtable Jordan Crenshaw - Senior Vice President, U.S. Chamber of Commerce This is the most contested AI policy conversation in Washington right now. The administration's executive orders rescinding Biden's AI EO, the push to preempt state AI laws through the reconciliation bill, the dismantling of AI safety review processes, and the deregulatory posture toward Big Tech are generating active legislative conflict. This panel asks the defining question of 2026: who benefits when Washington steps back from AI oversight — and who bears the cost? Panelists represent the full spectrum: a former federal regulator who knows what is being dismantled, a civil-society leader on what deregulation means for communities, a competitive-district Republican navigating the Trump AI agenda, a think tank scholar on U.S. vs. global regulatory divergence, and a veteran Silicon Valley legislator delivering a 30-year perspective on what this moment actually means. [Source](https://www.youtube.com/watch?v=kTRg8A6lcZU)

Source quotes

[0:04] Trooper Sanders: 简短的回答是，极其彻底地不管用（spectacularly no）。 [0:04] Trooper Sanders: 大多数参会者都是搭飞机来的，但谁敢坐一架毫无安全监管的飞机？AI 的复杂与危险程度比民航飞机高出指数级，必须建立确定的监管规则。 [5:45] Trooper Sanders: 如果你们的责任保险公司明天停止向你们提供承保，在座有几家企业还能维持运转？ [5:45] Trooper Sanders: 没有保险，现代经济体制就根本不复存在。AI 固然充满创新与颠覆性，但保险业可不是供你玩颠覆的行当。 [11:31] Jordan Crenshaw: 碎片化的“拼贴式”（patchwork）监管客观上只对资源雄厚的大公司和既得利益行业巨头有利，因为只有它们才有能力承担高昂的合规成本。 [11:31] Jordan Crenshaw: 你们做过与现行法律的冲突审查吗？查过这些问题是否能直接用现有州法解决吗？ [13:58] Trooper Sanders: 美国确实需要聪明的联邦法律、自愿性行业规范以及国际标准，但各方应当放平心态（"using our inside voices"）、扎实做好功课，由国会、白宫与各州州长展开持续的理性沟通与协作。 [17:19] Jordan Crenshaw: 数据中心实际上是公众对人工智能潜在风险在物理现实层面的直接具象化映射（"manifestation physically"）。 [17:19] Trooper Sanders: 美国的建国先贤们将政府体制设计得“缓慢而令人沮丧”（slow and frustrating），AI 监管不可能在几周内仓促强推过关，必须广泛吸纳企业及各方意见。 [17:19] Jordan Crenshaw: 监管流入 AI 系统的训练数据是治理的基石，众议院目前正在审议的《数据安全法案》（Secure Data Act）等具备优先权的立法应当率先通过，以奠定第一块制度基石。

Source quotes

[0:04] Trooper Sanders: spectacularly no [0:04] Trooper Sanders: exponentially more complicated and dangerous than airplanes [0:04] Trooper Sanders: student who didn't study for the bar exam until the night before [0:04] Jordan Crenshaw: to be determined [4:12] Trooper Sanders: dead on arrival in court very fast [6:46] Trooper Sanders: blinking yellow [6:46] Trooper Sanders: AI may be innovative and disruptive. You know what's not? Insurance. [12:24] Trooper Sanders: We need to have a consensus that government matters. [17:19] Trooper Sanders: infinitely more complex

Original transcript

[0:04] So, we have a great session which is AI under Trump deregulations, and we have got [0:10] two great panelists. And we're going to do the same thing. Have you introduce yourself [0:15] and basically, you know, answer one question basically. Um, and the, you know, setting the [0:23] stage for this is the federal government, and we've heard a lot of panels talk about this has [0:28] launched an effort to preempt state laws, and that's going to kind of determine how, [0:34] you know, you get AI, how does your AI get impacted either through the state or the federal [0:40] government. So, the question obviously is for both of you before you introduce yourself and answer is [0:47] Washington's approach to regulation working? And we'll start with you, Trooper. Introduce yourself [0:53] and answer. Sure. Thank you for having me. I'm Trooper Sanders, president of a new organization, [0:58] the State AI Safety Roundtable, that's trying to harness the power and potential of states [1:05] to create a better AI future for all of us. I think the short answer is spectacularly no. [1:12] The current approach that Washington is taking to AI regulation is is not working, [1:17] regardless of which side, frankly, of the issue you fall on. The state preemption so far [1:24] fell on its face, and I think, frankly, created a bipartisan coalition of governors, [1:29] attorneys general, and others who said, "Wait a minute. You you can't tell us that we can't do [1:34] anything for the next decade on AI." And so, I think it forced some bipartisan conversations. [1:41] I think the second part, and your previous panel touched on this, this is extremely difficult [1:47] issues to tackle. In AI, you know, most of us probably flew in here today. [1:52] How many of us would get on a plane that was completely unregulated? We wouldn't. [1:58] AI is exponentially more complicated and dangerous than airplanes. So, we are going to have to create [2:05] some regulatory certainty around this. And right now, um Washington has a lot of work to do, and [2:11] frankly, it's acting like the student who didn't study for the frankly the bar exam until the night [2:16] before because we're seeing just this week, just today or this week, the Trump administration's new [2:22] AI security agreement with industry. Well, that's regulation. But, we should have been [2:27] working on that over the past year than rushing in a matter of weeks to put something together. [2:33] Uh in a way, no is what uh Trooper meant. Jordan, is it working? Uh I would say uh to be determined. [2:43] But, uh I what I would say in terms of both Republicans and Democrats uh in in in Washington, [2:49] there is a fundamental misunderstanding uh about uh even what AI is. You know, we're just hearing [2:56] reports actually even talk about reports out of Florida today where there's misunderstanding [3:00] about what the definition of AI is in a new rule coming out of Florida potentially. [3:04] Um but, I think there's also a fundamental misunderstanding of the impact of AI. Rules don't [3:10] just impact large developers. They impact small businesses. Uh 65% of small businesses are worried [3:18] about a state patchwork. 58% of small businesses are using AI, and of those small businesses, 82% [3:25] are actually growing their workforce. They're worried about the regulatory uncertainty that's [3:29] going to come along with a patchwork of rules or bad rules that come from DC that are already [3:34] are really just discussed around a narrow set of companies. Um and I think the other area, too, is [3:41] Congress does not understand that AI is already regulated in many ways. Uh and uh [3:47] there is a regulatory sweet spot to hit. We don't have to reinvent the wheel, and and some have come [3:53] around to that. But, that that realization has not dawned upon all of Congress at this point. Uh, [3:58] introduction to a brief I I I'm Jordan Crenshaw. I'm senior vice president of the US Chamber's [4:03] Technology Engagement Center. So, Chamber is the world's largest business federation. We represent [4:08] everyone from the largest tech developers all the way to the smallest coffee shops on Main Street. [4:12] Okay, since you introduced yourself, we'll ask you. The Chamber has pushed for one federal [4:18] national law. Is there one state law that you think is, you know, hurting uh, [4:25] American businesses from an AI standpoint? We'd have said Colorado uh, up until uh, [4:33] a couple weeks ago, but I think, you know, the law that's still on the books right now that I [4:38] think is really concerning to us uh, is uh, the CCPA. It's the California Consumer Privacy Act. [4:44] Um, and its implementation through regulators. Uh, and one example of that is uh, the the Cal Privacy [4:52] Agency took one sentence of the CCPA, turned it into a 10-page, half-billion-dollar rule. Uh, [5:00] the federal government considers a hundred million a hundred million dollars to be a major rule. This [5:04] now quintuples that uh, in terms of its impact on interstate commerce. And and in fact, uh, [5:10] the Cal Privacy Agency even admitted that it would cost small businesses $16,000 annually to comply [5:17] with this broad set of privacy and AI rules uh, that they promulgated. Rules that even the author [5:22] of the CCPA, Alistair MacTaggart, questioned if they were actually legal when they voted on them, [5:27] but it was one of those, let's vote something out to see what happens, uh, kind of moments. And so, [5:32] I think that's the concern is that we are seeing uh, the rise of of states uh, trying to regulate [5:39] quickly in this space and not understanding the impacts, particularly on small business owners. [5:45] Trooper, you've written that uh, President Trump's executive order on preemption would [5:51] be dead on arrival in court very fast. Why do you say that? Well, I I think first of all, there are [5:58] there to say that states can't do anything flies in the face of the founders and the Constitution. [6:04] There are many areas that the Constitution gave states the power over our health and safety, the [6:10] the policing power. Um and one to not even have an explicit carve out around that would would be a [6:16] challenge. Um the the second is that federalism is a contested area. Often we have to battle that out [6:23] in the courts. And so if you overreach, usually preemption comes when Congress has passed a law. [6:30] And then there comes preemption. But to say before that that states can't do anything, [6:34] I would imagine you would see uh probably 50 states rush to the courts to get that struck down [6:39] because that is that is a danger to our republic and the way that federalism is supposed to work. [6:46] So if let's [snorts] say that uh preemption bills goes through, what harm do you think would [6:52] happen to citizens or businesses uh Trooper? You know, I I think it depends what what's [6:59] wrapped around the preemption. If preemption comes with with sensible federal legislation [7:05] that at least establishes a floor, you know, I think we can we can debate that. [7:11] I think if it is preemption uh in a way that that that frankly hands industry a free reign and hands [7:19] labs and others a free reign to do what you want, I think that would be [7:23] very dangerous. And let me give you a very concrete example. How many of you in here [7:29] could run your businesses or your profession if your liability insurer stopped covering you? [7:36] So one of the issues that commercial insurers are blinking yellow on is they [7:40] might not be able to cover frontier AI. When you're talking about a genetic AI, [7:46] when you're talking about things like recursive self-improvement and AI improving upon itself, [7:52] insurers are saying, "Wait a minute. We can't understand the risk. We can't quantify it. [7:58] We can't contain it. So, we may not be able to pay out." If we don't have an economy with insurance, [8:05] we don't have an economy. AI may be innovative and disruptive. You know what's not? Insurance. [8:11] And so, if we preempt states from dealing with sensible regulations, frankly, if we don't have [8:17] Washington create productive federalism with the states, it might not be a political matter. It [8:23] could just It could become, you know, small and medium-size and large companies suddenly [8:27] finding that they no longer have insurance to use AI. Now, we all know that the AI labs have [8:33] an enormous amount of money to make just to stay in business. If most of their customers, [8:38] which except for defense are highly regulated at the state level, if they can't get insurance or [8:43] their premiums jack up three, four, five times, they're not going to have much of a market. [8:50] Jordan, uh the chamber represents companies of all sizes, right? Small, medium, large, etc. [8:56] And you the chamber obviously pushes for deregulation per se. Do you think uh [9:04] with the AI uh you know, if we have this patchwork or bills, etc., it's going to hurt the startups [9:11] and small-size businesses? I I I I would say I Our position on AI is not necessarily deregulatory. [9:18] You know, we put together a commission a few years ago. Uh former Congressman John Delaney [9:23] and former Congressman Mike Ferguson led it. It was bipartisan. And they came up with a framework [9:29] that we think is flexible and still works. Is that First, we need to identify the harms [9:34] that we're trying to solve for. Second, are there laws on the books that can currently handle that? [9:41] So, we have every state in the country has a consumer protection law of general applicability. [9:46] Every state in the country has a civil rights or anti-discrimination law of general applicability. [9:51] Um 25 states now have a comprehensive data privacy law on the books. [9:56] We like to see one federal standard there, but if 25 others did that, [10:00] we we'd be supportive there, too, if it looks like the consensus model that's out there. So, [10:05] if if laws already exist, let's use those. Let's not reinvent the wheel. Then, the third step is [10:10] this: where has the law not caught up with the technology? Fourth step, develop risk-based [10:16] approaches to how to do that. And so so so that really is the appropriate mechanism. So, that's [10:22] things like passing the Take It Down Act, which went into law that that dealt with criminalizing [10:26] deepfake revenge images. Um that is uh you know, looking at potential frontier uh regulation. [10:34] But, at the same time, we don't have to reinvent the wheel and create completely new standard for [10:38] artificial intelligence when those standards work in discrimination, privacy, and consumer [10:42] protection law already. And so, one of the reasons why it's so difficult for small businesses [10:48] to compete is that they have to navigate this entire patchwork of state laws that's out there. [10:54] And we've done some research. We actually gave them uh uh in our Empowering Small Business [10:59] report we release every year, last year, we gave them the Colorado law. And we asked them what [11:04] what what's likely to happen if this goes into effect. About a third of them said it's going to [11:08] increase the cost of AI. A third of them said that they're less likely to even use the technology. [11:14] So, this means adoption goes down when we start seeing overregulation on the part of [11:19] certain states. You know, we estimated um that if Colorado's law went into effect, it would actually [11:25] have a $50 billion impact on the national economy and around 700,000 jobs that would be impacted [11:31] by just having Colorado's law alone. And And so, when you're a small business who is using [11:38] AI to grow and compete, you're using it to cut your cost to get off the ground and running. [11:43] Well, then when you have to start hiring a team of lawyers and compliance analysts to come in, [11:47] that negates the whole reason that you got AI in the first place. And so, regulation really only [11:53] helps incumbents and larger companies if it's a patchwork because it's more difficult to comply. [11:58] It's why we support a national framework and a national standard when it comes to both data [12:02] privacy and artificial intelligence so that we actually have one set of rules across the board. [12:08] But, there is a role to play for states. There are laws of general applicability that make sense [12:13] that can be used now that we want to see used. And so, I think in the long run, [12:19] it makes sense to have one standard so that small businesses have the ability to compete. [12:24] So, Jordan, I made this point in I think a few panels before. The governors of the states that [12:31] some of the states that you're citing say, "We won't have to come up with these laws [12:36] if the federal government would act or national standard or whatever there is." [12:42] So, what Are you saying that states should just stand down and wait for the federal government [12:48] to kind of get its act together in the next 3-4 years? What should they do? I think they should [12:52] take the framework that we've put forward. Is that look into your own research on what laws you can [12:59] use now to address artificial intelligence harms. There many cases where we've gone to [13:03] state legislators who are proposing comprehensive legislation, for example. We actually asked them, [13:07] "Have you run a conflicts check on current law? Have you run a check to see if this actually could [13:12] be solved with state law?" A lot of cases, no. And so, so I think that's why it's so important [13:19] that that that we have technology neutral laws in place that states can use [13:25] to regulate and actually enforce against harms that are caused by this technology. So, [13:30] there is a role for them to play that's already there, that's already been passed. And it's a it's [13:34] actually a good approach as well, particularly on the consumer-facing harms, because it means that [13:40] the legislature doesn't have to come back and pass a new law for a new kind of technology. So, [13:43] say when quantum hits, you know, we already have laws in place to address consumer harms that come [13:48] from that. We have laws to address discrimination laws that come out there. So, that's that's why [13:52] having these generally applicable laws makes sense to fall back on. So, Trooper, I mean, [13:58] what he's saying and some of I've heard some of the members of Congress say, "Hey, we've got the [14:02] FTC to protect. We've got, you know, the FDA to protect with medical devices. And we have so many [14:09] other institutions to protect. We kind of have it covered." Do you think we have it covered? [14:15] No, but I I think, you know, I would I would agree that we have a we have an enormous num- I think we [14:20] have an enormous potential Well, we want to record that both of you are agreeing on Yes. [laughter] [14:26] And and I also agree that, you know, we we have institutions. We have laws. States have a lot of [14:30] laws. I think the real issue is are they effective and do we have the muscle to make them um uh uh [14:39] effective? And I think, frankly, that comes down to we need to have a consensus that government [14:47] matters. So, we can have all the laws we want at the state level, which is important, [14:52] but if you have agencies that can't enforce them, that don't have the expertise to understand, well, [14:59] how might agentic AI apply to our decades-old licensing of nurses? When suddenly the hospitals [15:07] in the clinical setting, it might not just be nurses and doctors, but nurses, doctors, [15:12] and agents working together. If you if you're understaffed, you can't keep people safe. And [15:19] I think with that and things are going to happen. and then what that does, it poisons the politics [15:25] of AI because I think we would agree the last thing you want is something to happen. There's [15:30] a populist backlash and politicians are saying I have to do something fast. Cuz often when you [15:36] know, action in an emergency leads to bad policy. So for example, the Federal Trade Commission, [15:41] it should be a very influential and powerful organization. It has been purposefully kept weak [15:48] for decades. That doesn't work in the age of AI. If you take medical devices, um today one of the [15:55] founder co-founders of Google just left Google and is creating a new recursive self-improvement AI [16:01] company to automate scientific discovery. So that could mean that AI is coming out with all sorts of [16:08] scientific advances. Well, those things need to be approved by our regulators, be it a new chemical, [16:15] a new medical device, etc. They can barely do their job today with people doing it. [16:20] Imagine new processes, new discoveries, new products coming out by the weeks. It will collapse [16:27] and it will put the public in danger. So I completely agree that we should only pass [16:31] new laws where they are needed, but we need to strengthen all of our institutions to make [16:37] sure that in a common sense way government works. So you don't think we need a federal [16:44] regulation? [clears throat] Oh, I think I think we absolutely do. I I think we need a I think we [16:48] need a a smart law or perhaps series of laws. I think we need to encourage voluntary engagement. [16:54] I think international standards and other things, but you know, I but I think I like to think this [16:59] is an issue where we should be using our inside voices and doing our homework. And this is where [17:06] the Congress, the administration, and governors should be coming and legislators should be coming [17:11] together in a constant series of conversations and debates about how to how to get this right. [17:19] Jordan, I'm going to just ask a question. Has the chamber been surprised by this, so to speak, [17:27] uprising against data centers that has happened so rapidly in the United States, in our country? [17:35] No, I mean, I think we've seen legitimate concerns out there from from I would say authentic concerns [17:42] coming from communities. It's concerns about new facilities going in people's backyards. It's [17:46] facilities, you know, the issue about ratepayer issues that we hear in the news every day. These [17:51] are authentic concerns that we're hearing from from citizens. I think one of the things we've [17:56] seen though is there's been attempt by some of our outside competitors to leverage and use [18:02] those concerns to actually inhibit our ability in the US to compete on AI. And and so I I I think [18:09] one of the things I think I am surprised about is how quickly this has happened, [18:13] but I'm not surprised that these arguments are coming up considering all the uncertainty that's [18:17] been generated around AI itself. Um I think it was heard earlier in panels earlier today was that, [18:23] you know, these data centers are the manifestation physically of some of the concerns others are [18:27] having about artificial intelligence and it it to some extent does make sense that that's why [18:31] we're hearing this now. Trooper, you worked in the White House advising them on domestic policy. How [18:39] does this regulatory, so to speak, battle compared to some of the others that you have seen? How [18:45] would you compare them to this? I mean, I would say it is infinitely more complex and they are [18:52] choosing to give themselves an extraordinarily small amount of time to get this right. One of the [18:59] areas I worked on in the 1990s was around mental health parity, that your mental health care should [19:04] be covered equally in as your physical health care by your health insurers. That took years and years [19:11] to work through substantively, politically, you had to do revisions to get the to get it right. [19:17] Just today Congress there's a new bill to regulate an obscure prescription drug pricing regime that [19:25] took years to work through. You you can't take something like AI and AI safety and ram it through [19:32] in a few weeks. You need the chamber. You need to hear from businesses and others to get this right. [19:38] I mean, you know, look the the founders constructed our government to be slow and [19:43] frustrating. And so we have to make sure that AI can fit within that. So we have to get to work. [19:49] Final question for both of you because we have our friend out there. Will a federal preemption bill [19:57] pass through Congress ever? Tooper, yes, no? Oh, absolutely. I I I actually think there's [20:03] hope it will it will be done in the I don't think it'll be done this Congress, but I think in 2027. [20:08] Eventually. But I think the most likely and I think the most urgent at this point is [20:12] comprehensive privacy legislation because you need to regulate and govern the data that goes into AI [20:17] systems. That's why we need to see something like the Secure Data Act which is in front of [20:20] the House of Representatives right now pass as a preemptive measure to have that first building [20:24] block in place. Thank you to both of you for such a great conversation. Unfortunately, you agreed on [20:31] almost everything, but we'll try to fix that next time. Thank you so much. Thank you. [applause]