In this photo illustration, the logo of open-source AI agent OpenClaw is displayed on a smartphone screen on March 9, 2026 in Beijing, China. (Photo by VCG/VCG via AP )


Building Trust Infrastructure for Agentic AI

A rock embedded with intricate circuit board patterns, held delicately by pale hands drawn in a ghostly style. The contrast between the rough, metallic mineral and the sleek, artificial circuit board illustrates the relationship between raw natural resources and modern technological development. The hands evoke human involvement in the extraction and manufacturing processes.

Rare Metals 2 by Hanna Barakat & Archival Images of AI + AIxDESIGN / 

Without Open-Source Hardware, There Is No EU Tech Sovereignty

On the left is a grey-scale version of a rat with inaccurate and oversized reproductive organs recognizable from an academic article which contained AI slop. The image has been edited to be sliced, and blue painted torus icons overlay the image. The background features a green mountainous range with a magenta tiled floor and gradient sky. 

AI Didn’t Kill Design—It Exposed It

Detail from Digital Society Bell by Lone Thomasky & Bits&Bäume. 

How State Utility Regulators Are Shaping AI Infrastructure Deployment

European Flags in front of the European Commission Headquarters building. 

Brussels, Belgium - November 3, 2022; European Flags in front of the European Commission Headquarters building. 

Designing Europe’s Search Data Sharing Rules for Competition in the AI Era

The exterior of the United States Department of Commerce in Washington, DC.

Alex Stamos on Why the US Should Lift Its Fable and Mythos Export Ban

President Donald J. Trump holds a cabinet meeting in the Cabinet Room Wednesday, May 27, 2026. (Official White House photo by Daniel Torok)

President Donald J. Trump holds a cabinet meeting in the Cabinet Room Wednesday, May 27, 2026. (Official White House 

Transparency and Accountability Gaps in Trump's New AI Executive Order

This features two images, the background image is a road with a zebra crossing. Overlaid is an image of a typical street with buildings either side and cars lined up on the side of the street. There are also trees and a large stretch of road which centers the image. The whole image has a purple filter and a pixel effect. 

AI, Privacy, and the Hidden Architecture of Harm from Inference

AI Am Over It by Nadia Piet & Archival Images of AI + AIxDESIGN / 

AI's Catastrophic Risk Isn't Rogue Machines, It's Cognitive Surrender

The image depicts a large, diverse crowd of miniature figures, representing a wide variety of people. The figures are painted in different colors, wearing various clothing styles, and some appear to be engaged in conversations or interactions. The focus of the image is on the middle part of the crowd, with the surrounding figures gradually blurring into the background, creating a sense of depth. The diversity of the figures, in terms of skin tones, attire, and activities, suggests themes related to social cohesion, diversity, and community.

Toward Representative AI Governance

A Meta sign at Meta Headquarters in Menlo Park, California, in January 2025.

Tech Workers Are Fighting Against Silicon Valley’s AI Push

3D image photo collage template sketch of an anonymous spyware business.

Canada Has a Precedent to Regulate Surveillance Pricing

Protesters march through downtown Memphis, Tennessee, carrying signs opposing Elon Musk and xAI during the "Get Out of Memphis" demonstration on October 4, 2025. (Photo by AUSTIN JOHNSON/AFP via Getty Images)

Addressing Regulatory Arbitrage in the AI Supply Chain

Brazilian President Luiz Inacio Lula da Silva holds a cabinet meeting at Planalto presidential palace in Brasilia, Brazil, Wednesday, June 3, 2026. (AP Photo/Eraldo Peres)

Big Tech's Proxy War Against Brazil's New Platform Accountability Rules

Members of the European Parliament Arba Kokalari and Michael McNamra pictured during the AI omnibus interinstitutional negotiations. / Photo: EP

Members of the European Parliament and Arba Kokalari and Michael McNamra pictured during the AI omnibus interinstitutional negotiations. / Photo: EP

EU Lawmakers Move to Ban AI Nudifiers, but Enforcement Remains Unclear

An array of colorful, fossil-like data imprints representing the static nature of AI models, laden with outdated contexts and biases.

Luke Conroy and Anne Fehres & AI4Media / 

A Case Against Forever: Why Cloud and Social Media Need Inactivity-Based Deletions

Howard Lutnick, US Secretary of Commerce, testifies before the House Committee on Appropriations Subcommittee on Commerce, Justice, Science, and Related Agencies in the Rayburn House Office Building in Washington, DC on Thursday, June 5, 2025. (Photo by Aaron Schwartz/Sipa USA)

Did the US Government Just Set An AI Export Precedent by Blocking Mythos?

An abstract image based on the US Constitution

Earning a Social License for Transformative AI

BRUSSELS, BELGIUM—JUNE 3: EU Commissioner for Tech Sovereignty, Security and Democracy Henna Virkkunen introduces the European Technological Sovereignty Package, featuring the Chips Act 2.0, Cloud and AI Development Act, and related strategies to reduce the EU's dependence on non-European suppliers for critical digital technologies. (Photo by Thierry Monasse/Getty Images)

Europe’s AI Sovereignty Problem Runs Far Deeper Than Frontier Access

The Claude Mythos logo is displayed on the screen of a smartphone placed on a reflective surface.

The Claude Mythos logo is displayed on the screen of a smartphone placed on a reflective surface onto which the promotional image for the model is projected, in Creteil, France, on June 10, 2026. Anthropic announces the release of two new Mythos-class artificial intelligence models designed for cybersecurity and biomedical research, targeting both consumers and businesses. (Photo by Samuel Boivin/NurPhoto via AP)

The Mythos Recall and the Global AI Sovereignty Debate

Chris Riley is the executive director of the Data Transfer Initiative.

In this photo illustration, the logo of open-source AI agent OpenClaw is displayed on a smartphone screen on March 9, 2026 in Beijing, China. (Photo by VCG/VCG via AP)

“Agentic” is unquestionably the buzzword of 2026. Startups and tech behemoths are racing to build the most hands-off technology futures possible, even as public and policymaker response remains muted where not overtly hostile. The result is a substantial governance debt—and the currency at stake is trust.

I have oriented my professional career around advancing the open internet. A formative portion of it was on net neutrality, ensuring that telecom operators do not discriminate in the quality of packet delivery based on the source or service of the data. I’ve also worked on intellectual property, internet censorship, competition, and a variety of other issues that relate to whether or not people are empowered to create and use technology and the internet as they see fit, without any undue friction.

But “open” can also mean “risky.” At the network layer, routers and other devices have security filters that can catch some harmful traffic, but not all. At the device layer, operating systems do their part as well, and within that, app stores and other pathways by which users select the software and the tools to which they trust their online experiences and their personal data. Privacy and security can absolutely go hand-in-hand with openness, but it takes a little of the right kind of friction, and a lot of behind-the-scenes work and infrastructure to establish and maintain trust.

The AI landscape today feels remarkably open, particularly when compared to many other technology paradigms at similar points in their history. Market shares are not set in stone, competition is intense, and the paradigms and value propositions change in the blink of an eye. Intermediaries allow multiple models to be used within the same frameworks and user-facing platforms, for a range of purposes. And while there’s still a hot news cycle whenever a frontier lab drops its latest model with updated leaderboard rankings on various performance metrics, the greater narrative has moved beyond the engine itself to the surrounding ecosystem: the agentic universe.

 that the core engines of AI would increasingly become commoditized, and what would matter would be who had access to users’ personal data, increasingly the key to value generation. Three years later, I am certain this is the case. And the work of the organization I lead, 

, remains focused on empowering people to use their data safely and securely where they choose—in all contexts, 

The agentic shift of the AI conversation has shined a spotlight on the value of personal data for an audience far bigger than the usual portability crowds. It’s technically and practically different from the original generative AI conversation, where bulk training data and massive models produced the key value. Now, what matters is less the model you run (though it still matters), and more whether your agent can manage your finances, your email, your digital life that determines whether it is useful to you, or whether it becomes just another icon on your phone that you tried once and abandoned.

One tool in particular, OpenClaw, blew the doors open on whether people would want to use agents in their day-to-day lives. The project started as a proactive digital assistant, a tool you run and give access to your services (and thus, crucially, your personal data) and ask to act on your behalf. It quickly became the most starred project in GitHub, and earned the backing of leading AI organizations OpenAI (which 

 OpenClaw creator Peter Steinberger) and 

. Millions of people began using it, generating what many regarded as more tangible benefits from AI than any prior tool.

As significantly, OpenClaw’s emergence showed that incredible value can be created in the agentic AI ecosystem by anyone. Unlike many other technology contexts, vertical integration or deployment within a vertical stack run by an established, major company is simply not needed. The data is already accessible to the user through their machines, and the models are offered via APIs (not to mention the rich library of open weight models on 

). All it takes is handing over all your digital keys to a piece of software you just downloaded off the internet.

What we are beginning to learn today, and will continue to learn in the weeks and months to come, is that behind this wonderful open dynamic lies a broken governance system. It’s not a matter of a single weakness, although there are plenty of individual examples that can be cited. Rather, 

: “What makes OpenClaw so dangerous is not a single exploit. It is the collapse of data governance boundaries across the entire AI agent lifecycle.” Or 

: “Because OpenClaw requires deep integration into your messaging and file systems to be useful, you are essentially building a high-speed bridge for malware.”

It’s the combination of deep access to personal data and an open third-party ecosystem that makes this particularly alarming. Researchers looking at the ClawHub repository of OpenClaw skills and broader ecosystem agentic repositories have found 

 to be malicious—not merely insecure, but intentionally harmful. Little wonder security researchers at firms like Cisco call it “

And our legal system is not nearly equipped to step in. In several countries, including the European Union and United States, I see increasing awareness of security and safety considerations in models themselves. But agents are something else entirely. In the EU, the AI Act’s Article 73 requires incident reporting for high risk incidents; as a Tech Policy Press piece noted, agentic AI constitutes 

,” articulates five dimensions of shortcomings in the AI Act facing the agentic future: performance, misuse, privacy, equity, and oversight.

The agentic floodwaters are rising, and water is leaking through. How can we build a durable dam of greater trust? If we can’t create powerful enough screens and filters to identify and tag the harmful parts of the ecosystem—which seems like more of a certainty than an ‘if’—then our alternative is to whitelist those that we can trust.

The concept of trust in the AI ecosystem is complex and nuanced. The George Washington University professor David A. Broniatowski articulates 

: technocratic, in which trust is a metric to optimize; relational, emergent through social cohesion and structure; pragmatic, built on institutions and established practices; and critical, viewing trust through the lens of political and/or economic power. Through this lens, the trust gap in agentic AI I’ve been describing is pragmatic. After all, relational trust seems present with public willingness to try OpenClaw and similar technologies in the absence of metrics or infrastructures or laws that protect their interests.

: boring, ordinary, transparent, effective institutions that test and validate trust on behalf of people, and create a world where normal users of agents don’t have to think or worry about it, and can take advantage of the upsides of an ecosystem with a baseline of protection from the concomitant risks.

That takes a lot of pieces. And there are efforts working on some of them. For example, 

 promises robust and unique identification of agents, and tools to ensure they are fully and validly authorized by users to take actions.

An occasional (though some might say persistent) myopia within the technology world is an assumption that the special circumstances of the present moment dilute, or defeat, lessons from past practice—that everything is new and a greenfield solution is called for. What, really, is different in the agentic AI context in terms of trust and the flow of personal data? Speed and reduced friction of development, deployment, and adoption is a clear change. But the abstract architectures and responsibilities are the same. Someone, person or entity, produces and ships a piece of software; that software communicates over the internet to a source of personal data; and a user authorizes the source of that data to make it available to the software, which uses it in some manner, including potentially passing it along to other software.

We need to be able to identify this processing software and ensure that it is properly authorized by a user. And we need to be able to ensure that the ways in which it handles that data once received are acceptable. We need to know the entity behind the software and how it built the software’s data processing—what its privacy policy is, what its security methods are, how it communicates through the software or otherwise to the user what it will do with the data.

These are the core questions behind DTI’s 

. We built DTR for the context of data portability, which is nominally orthogonal to the agentic AI ecosystem. But more and more, we are seeing in agentic AI emergence of the same challenges of trust as in any software or services that handle personal data at a user’s request. And we see the same need for public, open, transparent, accountable infrastructure to establish and maintain trust. So let’s build the dam we need, to protect people and to set agentic AI up for success or failure on its capabilities rather than its most obvious risks.

Google’s dominance in search is increasingly under scrutiny as regulators on both sides of the Atlantic consider whether new rules could reshape competition in the AI-driven search era. Against this backdrop, earlier this year, the European Commission initiated 

 regarding Google’s noncompliance with 

 of the Digital Markets Act (DMA), which requires the company to share ranking, query, click, and view data with its rivals.

A public consultation on the Commission’s 

 setting out how this data-sharing obligation should be implemented in practice recently concluded. In 

submitted during the consultation period, the Knight-Georgetown Institute noted that the Commission’s proposals represent a comprehensive, well-targeted, and AI-responsive first step toward search data sharing, with the potential to unlock what may prove to be a key ingredient for stimulating competition in the AI-powered search era.

To strengthen the proposed framework, the Commission will need to resolve ambiguities in data scope (particularly the treatment of AI-enabled features) and sharing frequency, specify explicit privacy and utility goals to govern the design of anonymization measures and ongoing evaluation by independent experts, and ensure access to a dispute-resolution process and negotiation flexibility for data recipients.

Mandatory asset sharing has a history of unleashing innovation in digital markets. In 1956, AT&T, then a leading company in computing, agreed to a 

 with the US Department of Justice requiring it to license its complete portfolio of patents (including key technologies like the transistor) royalty-free. This sharing remedy was crucial for seeding the 

 that defined the early computing industry. 

 has since found that AT&T’s patent licensing was responsible for a 12% increase in innovation (relative to non-licensed patents in the same technology class) and a substantial increase in overall innovation.

Seventy years later, we are in the midst of another computing revolution, where mandatory asset sharing may prove critical once again. In addition to the Commission’s actions, the remedies issued last year in the landmark 

 to share a substantially narrower universe of search-query data with its rivals. Although the case is 

, the formation of the court-ordered Technical Committee (TC) to assist with implementing the remedy package is ongoing. 

If and when these search-data-sharing regimes come into force, the critical questions become whether and how they can best catalyze competition in online search while protecting user privacy. 

A central design choice of any data-sharing regime is the scope of data that must be shared. The European Commission's preliminary measures take a broad approach, requiring Google to disclose comprehensive anonymized information about search results and how a user interacts with them. For instance, this might include the content of a key snippet of a webpage displayed in response to a search query, and how long a user’s cursor hovered over it before clicking on a link. This expansive approach to data sharing, paired with technical and contractual anonymization controls, will equip rivals with a key ingredient necessary to improve the quality of their own search products and compete more effectively with Google.

An important question hanging over enforcers is whether to fold in the generative AI features that increasingly 

 in how users interact with search products.

The Commission’s approach appears to go further than the remedy in the 

 case, which excludes user data that Google uses to train search-related generative AI models from the data-sharing requirement. The Commission’s proposed measures, on the other hand, appear to scope in AI features like AI Overviews. Google is required to disclose all “visual content” on the results page, which presumably include AI features such as AI Overviews and AI Mode outputs. The Commission should make the definition of what is included in “visual content” explicit so that it is clear whether AI-related features are covered.

The performance of machine learning models that power search products typically improves when search engines have access to “fresh” queries, as has been confirmed in 

. The Commission’s proposed approach lays a strong foundation for frequent data sharing in establishing the parity principle: rivals’ access to search data must be on par with Google’s. However, the Commission’s measures separately specify that Google must share its search data “daily.” It is unclear how this provision interacts with the parity requirement. The Commission should clarify this ambiguity in its final measures.

This is also an area where US antitrust enforcers could learn from the EU’s approach. The 

 leaves the data-sharing frequency question somewhat open, specifying that Google must share query data “at least twice,” with the exact frequency depending on the utility of the dataset after privacy-enhancing techniques are applied. The European Commission’s preliminary measures demonstrate how ambitious this data-sharing remedy could be; the TC could similarly conclude that requiring Google to share its search query data at a specified rate (e.g., once per day) is feasible and desirable.

Given the scale and sensitivity of the data involved, privacy safeguards will be critical to the success and legitimacy of the framework. In its final measures, the Commission should adopt an evidence-informed approach to assessing the relationship between privacy and competition.

The Commission’s preliminary measures include a layer of comprehensive technical protections intended to “mitigate the risk of re-identification of end users to a residual level.” The technical protections are thoughtful and innovative, but the Commission did not derive them from any clearly specified design goals. For example, the measures aim to reduce the risk of user re-identification to a "residual level," without specifying what the Commission considers to be an acceptable threshold. The Commission should also state any other goals it has for the technical protection measures as a whole, including for data utility or contestability. The set of such measures the Commission has proposed cannot be evaluated without knowing the goals of the system.

The Commission should explain the motivation and rationale that it used for the design of the technical measures, including the choice of each parameter specified in the technical measures. The technical measures appear to be broadly aimed at mitigating the risk of user re-identification while maintaining utility of the search data, but the proposed measures do not explain precisely why the Commission chose each measure nor the specific choices made in the design of each measure. The technical measures appear to have been designed with implicit preferences with respect to utility and privacy.

In addition, the Commission should establish a process for ongoing evaluation of search data sharing against its privacy and utility goals, preferably conducted by independent experts. As search behavior changes and more competitors obtain access to the shared data, performance against the privacy and utility goals of the technical measures may degrade or shift. User behavior is not static, especially as AI-enabled features spark change in the 

 received by search products, implying a need to regularly reevaluate how the technical measures are designed. Independent experts are best suited to this task because both Google and its competitors have vested interests in seeing the results of ongoing evaluation turn out in a way that affects the scope of shared data. Evaluation should thus be conducted by neutral expert parties.

This is another area where US enforcers can learn from the European approach. The Final Judgment is 

 gives the TC discretion to determine appropriate privacy and security safeguards for shared search data. The TC is potentially well-equipped to specify design goals through empirical derivation and provide for ongoing independent assessment of the efficacy of the safeguards.

While the Commission's preliminary measures are detailed, they still leave Google with discretion in several important areas: the processing of applications, pricing, the scope of contractual limits on data sharing, and the mechanics of the data transfer process. Given the pro-competitive potential of search data sharing, Google has incentives to 

 in its compliance efforts. Each area where Google has discretion can therefore create a friction point at which delay or disagreement can arise. Besides disputes over pricing, the preliminary measures do not yet envision how such disagreements would be resolved. The Commission’s final measures could reduce these frictions on the front end by establishing a robust dispute-resolution mechanism to serve as a backstop.

US antitrust enforcers face a similar set of questions about the data-sharing process. The 

 Final Judgment leaves the dispute-resolution process open-ended. The TC can play a role in addressing this gap. 

As the next chapter of competition policy in digital markets unfolds, the search data-sharing regime taking shape in the EU will rise or fall on how much data Google must share, how often, with what privacy safeguards, and through what process rivals actually receive it. The Commission has built a framework that, with the refinements outlined above, could serve as a model for jurisdictions around the world. Past antitrust interventions, including the AT&T consent decree, show how asset-sharing remedies can unlock innovation. 

If the search-data-sharing regime in the EU succeeds, it will not only shape the next chapter of search, but could also transform competition in AI-powered information access for a generation.

State public utility commission (PUC) regulators are now active participants in shaping the deployment of AI infrastructure. 

, 27 states have legislation pending to govern how new large loads (a category that encompasses most modern data centers) are evaluated, priced, and connected to the electric grid. The substantive questions these rules address are familiar to utility regulators, yet largely absent from AI policy discourse: Who bears the cost of new electricity generation and transmission investments needed to serve these facilities? How to assess whether utility commitments to new infrastructure are prudent given uncertain demand? And what regulatory protections should existing customers have against the risk that projected demand does not materialize?

 regulation socializes infrastructure costs across the ratepayer base, a framework workable for incremental load growth but increasingly strained when a single new customer can require investment approaching the scale of existing customer classes. 

, set to take effect in July 2026, offers a framework for understanding how this layer of regulation works in practice. The statute requires that any new load above 50 megawatts receive service only under a PUC-approved contract. That contract must satisfy a statutory “no harm” test and demonstrate that the new load will fund its full share of generation, transmission, substation, and distribution infrastructure costs. Many other states are considering legislation to address these same issues. Idaho’s statute is unusually clear in how it operationalizes cost-causation principles, making it a useful reference point for state frameworks evolving in response to AI infrastructure demand.

Who regulates AI-related energy infrastructure: the federal and state jurisdictional architecture

 states where a vertically integrated, investor-owned utility holds a monopoly franchise and the PUC sets retail rates, approves major capital investments, and reviews resource planning projecting future demand. PUC proceedings are adjudicative and adversarial, with utility filings tested through discovery and testimony from intervenors. The decisions are subject to judicial review.

Regulatory authority over these kinds of electric utilities is split between state and federal jurisdictions. Generation acquisition by a regulated utility for retail service is primarily a state issue. Utility plans to build or contract for new generation are reviewed through state Certificate for Public Convenience and Necessity (

) proceedings, which assess whether proposed projects are in the public convenience and necessity, and Integrated Resource Plan (

) processes, which evaluate long-term planning options. Approved costs are recovered through retail rates set by the state PUC. Wholesale electricity generation, including generation sold into competitive markets, falls under the jurisdiction of the Federal Energy Regulatory Commission (

For AI infrastructure served by traditional regulated utilities, cost causation questions are settled primarily at the state level. Transmission regulation is more complex: state PUCs have authority over siting and retail rates within their borders, while FERC has authority over wholesale transmission service, interstate planning, and the regional transmission process administered by Regional Transmission Organizations and Independent System Operators. FERC also issues orders affecting national transmission planning, including recent rules on regional planning, electricity generator interconnection reform, and long-term planning aimed at large load growth. Major new transmission projects often require both state and FERC approvals.

Where AI infrastructure lands within this regulatory architecture depends on scale. A data center that connects to a regulated utility's network typically triggers state regulatory processes only. Larger deployments requiring new high-voltage transmission or significant new generation commitments require coordination across state PUC processes and FERC regional planning. Multi-state transmission projects like the 

 line running from eastern Wyoming to Idaho and also serving PacifiCorp's six-state territory illustrate these challenges: each state reviews costs and benefits within its own jurisdiction while FERC oversees regional planning.

The regulatory pathway is rarely cleanly federal or cleanly state-based. Federal policy proposals focused on

alone miss the substantial role that state PUCs play in approving underlying generation and transmission investments, while state frameworks operate within constraints set by FERC's transmission planning and interconnection rules. Understanding either layer in isolation produces an incomplete picture of where binding constraints on AI infrastructure deployment actually lie.

What state regulators are deciding now will shape deployment for decades

Substantive disputes in these proceedings cluster around four interrelated questions, each being resolved in the current state PUC proceedings and legislative action.

States are actively defining which costs incurred to serve new large loads are properly assigned to those customers and which should be socialized across the entire rate base. The traditional cost-of-service model provides some guidance: costs that would not have been incurred but for the new load are typically directly assignable, while costs that benefit the broader system are typically allocated.

Applying that principle to AI infrastructure, however, remains contested. If a utility builds new generation that exceeds new demand from a data center but is justified by the planning reserve margin (the excess generating capacity above forecast peak demand that utilities maintain to ensure reliability) that it requires, who pays for the excess capacity? Idaho HB 911 takes a strong position: new large data centers are responsible for their full share of generation, transmission, substation, and distribution investments that would not be required by the public utility "but for the new large load." How that test is applied in individual proceedings will determine the share of infrastructure costs developers bear. Variation across jurisdictions will, in turn, shape where data center development is economically viable.

States are also working out, in real time, how the doctrine of prudency applies to utility resource commitments driven by AI-related demand projections. 

 asks whether an investment was reasonable given what the utility knew at the time of the decision.

Applied to AI infrastructure, the inquiry is novel. Projections for data center electricity use vary considerably depending on assumptions about the pace of AI deployment, and utilities making long-term resource commitments based on those projections face a real risk that load will not materialize at the projected scale or timeline.

The standard PUCs adopt now will shape utility behavior going forward, both in how aggressively they commit to new resources to meet speculative energy demands and in how those commitments are structured. Strict prudency standards push utilities to require firmer customer commitments; permissive standards allow faster buildout but greater ratepayer exposure to unrealized projections.

Stranded-cost risk and financial security

A stranded investment is one made to support infrastructure that no longer has a customer base sufficient to recover the costs. A utility that builds generation and transmission to serve a data center that later shrinks, relocates, or fails to obtain the required federal permits may be left with stranded costs traditionally recovered from its remaining ratepayer base.

 are now requiring financial security from new large-load customers to 

against this risk. Idaho HB 911 requires these new users to furnish financial security "in a form and amount approved by the commission, that is reasonably sufficient to protect the public utility and its other customers from the risk of stranded costs, unrecoverable costs, or unrecoverable investments."

The level at which PUCs set those requirements will determine how much skin in the game AI infrastructure developers must have before a utility commits.

The gap between projection and realization

Underlying these questions is the difficulty of regulating long-lived infrastructure investments based on demand projections that may never materialize. IRPs must project demand decades into the future, and AI-driven load growth has introduced a substantial new source of uncertainty.

 after a utility has already committed to new infrastructure on the basis of original projections. Other projects appear likely to 

State regulators are now developing mechanisms to manage this uncertainty, including

 that align the utility's obligations with the customer's actual demand schedule. The frameworks adopted now will determine how regulatory systems handle the next several years of AI infrastructure buildout, whether AI demand projections prove too aggressive or too conservative.

Where AI infrastructure policy is actually being decided

The state regulatory layer is a material variable in AI infrastructure deployment, and the frameworks governing it are being shaped through proceedings happening now. Decisions made in 2026 on cost causation, prudency, ratepayer protection, and demand realization risk will set the terms for where data centers can be built, what they will cost, who bears the financial risk, and how fast capacity can come online

For the AI policy community, the implications are practical: Federal AI policy proposals, whether focused on permitting reform, federal preemption, or industrial strategy, operate within a regulatory landscape substantially shaped by state PUC decisions and state legislation. Analysis that fails to engage with the state layer tends to overestimate federal impact, underestimating binding constraints. The substantive proceedings before state PUCs are not background noise. They are squarely in the foreground.

These proceedings are public: Filings, testimony, orders, and legislative records are all accessible. The questions at stake are familiar to utility regulators but are not yet familiar to most of the AI policy community. Closing that gap is a reasonable next step for anyone whose work intersects with the physical deployment of AI infrastructure.

of student essays produced in collaboration with the Berkman Klein Center for Internet & Society at Harvard University.

“The design process is dead. Here’s what’s replacing it.” This is the title of 

, the lead designer for Claude at Anthropic, that set the design world abuzz. Wen’s provocation intentionally strikes at the heart of what many product designers in the tech industry see as their core job’s purpose: to guide the direction of the products that they work on, through the design process, towards human-centered outcomes aligned with business goals. Without the design process, how do we make sure that our products are aligned with what people need?

As a designer, I can sympathize with many of my colleagues’ gut reactions to Wen’s provocation. Yet much of the discourse surrounding it fails to recognize that the adoption of AI is only making more visible the fact that the design process was never solely in pursuit of the benefit of people who use the things that we make. Power over how products are designed has always been in the hands of the companies that fund their creation, and AI makes explicit the ways that the design process is primarily used to advance company goals, rather than support the users it is meant to serve.

 to what design practitioners have always known: that the design of platforms themselves—not just the content they host—directly influences the negative behaviors that we see coming out of the use of tech, particularly on social media. The design process that shapes this form, popularized in corporate settings by 

 as a way of centering product decisions around human needs, has been a core value proposition of the product design discipline. Commonly used frameworks for the design process in the tech industry, such as 

, start from a human-oriented place: researching the impact of a problem on real people. This research is synthesized to form a better definition of the problem, spurring the development of multiple solutions that then get tested amongst users to ultimately determine the best solution. The business value proposition is that this process guides product innovation towards solutions that people actually want and use, while the value for social good is that the process is supposed to embed real people in the creation of technology.

Contrast that design process with this scene: an anonymous user types “make me a music app with a beach vibe” into a text box and presses a button labeled “generate.” Multiple screens representing an appropriate music app instantly pop up, as if the magic wand icon accompanying the button created magic that replicated the efforts of a designer. Words describe how the user can transfer this generated design into Figma, the most common product design tool in industry, or implement it directly into code. 

 announcing Stitch, Google’s newest AI tool aimed at streamlining the product design process. The instant, almost magically-created, design work that strays from the user-centered approach of the design process is typical of other recently released AI-assisted design tools, such as 

 of companies that create design software. One way to read this is that the market is signaling the future demise of the role of product designers, reckoning that their role can be replaced by AI tools that can do the same work faster and more seamlessly with engineering. The argument that Wen makes for this new way of working aligns with that: working hand in hand with engineers to quickly iterate, with AI-empowered engineering implementation driving product direction.

However, during the seven years I spent working in the tech industry as product designer for emerging tech firms before the rise of generative AI design tools, I found that instead of spending most of my time experimenting with new solutions to help connect people better through technology, I spent more time convincing others that doing good by the people using our products was aligned with business goals. On one product I worked on, it took months of self-driven user research and internal advocacy work to successfully advocate for women to be considered an integral part of the product’s target audience. On another product, I was unsuccessful in persuading management to find alternatives to 

-style mechanics and eventually the output of my work was an extractive gamification system that powered the product’s social dynamics. When I attempted to experiment with interactions and mechanics, it would frequently conflict with product roadmaps that prioritized monetization features.

As a designer who values doing good by the people that I design products for, those experiences made clear to me that the incentives of the tech industry often were at odds with how tech companies branded themselves as doing good in the world. 

 the idea that technology by itself does not solve social problems, but that it only amplifies human intent and capacity. That is, technology amplifies existing power structures and incentives, instead of acting as a solution to them. In line with this theory, I believe that AI has simply amplified the incentives that currently exist in the tech industry, and the benefits that AI companies tout as evidence of their ability to change the world only reveal the values that underpin how decisions are made within the tech industry.

Instead of waiting for designers to iterate on interfaces before engineers implement them, engineers themselves can spin up multiple designs on their own to test out ideas, shifting decision making power towards those who rush to implement. Designers get caught up in the demand for quicker solutions and learn to spin up prototypes in order to remain part of the decision making process. Where speed is prioritized because it provides companies with clear advantages—first to market, quicker time to release features that a competitor does not have—AI provides an answer.

However, these shorter development cycles also lead to shrinking windows of agency for designers to intervene in decisions that may serve company goals but run counter to what they believe is best for users, because there is simply less time to raise concerns or do the advocacy work needed to convince management of a shift in direction. The problem becomes compounded as less labor and company resources are needed to ship a feature. Because of the lower cost of shipping a feature, the default answer to a concern raised becomes “let’s see how it goes—we can easily change it later if it doesn’t work.”

Wen’s approach to ship fast and iterate publicly “building trust through speed,” is representative of those values and it is the newest embodiment of the old Facebook motto that the tech industry has baked into its culture: “Move fast and break things.” When designers, the ones tasked with advocating for users, lose the agency to intervene in product decisions, the consequences are passed onto the millions of users that rely on those products. Products become less responsive to their needs and more extractive in their design.

I see this already in my work. Before design and product teams agree on important requirements or understand what users really need, engineering has already spun up a prototype that meets their understanding of the problem. These prototypes lead the conversations about the product, as it is easier for people to respond to existing artifacts that they can interact with, even when they conflict with the actual user’s needs. Worse, I have been presented with engineering builds that embed assumptions that conflict with user needs into the underlying system architecture, making them much harder to change without a significant overhaul of the app itself.

While US courts are only recently grappling with the ability to hold tech companies accountable for their design choices on social media platforms, the product design field is already using AI to accelerate the potential to create even more design paradigms that negatively impact users that courts have yet to contend with. I do not see this as inherently a problem with AI—it is simply amplifying the existing problems in the practice of product design. AI is not going to make companies value design if they did not value it to begin with, nor will it make companies care about user edge cases or the negative impact of their product on people’s lives more than it did before. That would require a shift in the tech industry towards valuing the importance of thoughtful design. More importantly, it would require a paradigm shift towards an industry-wide questioning of the values of speed, productivity, and profit that underpin how the tech industry builds and evaluates its products. More fundamentally, it would require a reconsideration of who should have the power to shape the systems that millions of people rely on.

Helen McEntee, Ireland's Minister for Foreign Affairs and Trade, outlines priorities for Ireland's upcoming EU Council Presidency on June 10. Source: Irish Presidency of the Council of the EU 2026.

Ireland Targets AI Omnibus, Child Safety in EU Presidency Agenda

Abhijeet Dipke, center, founder of the Cockroach Janta Party addresses supporters during a protest demonstration in New Delhi, India, Saturday, June 6, 2026. (AP Photo/Manish Swarup)

After Test Data Scandal, India’s Youth Stake Their Claim on the Digital Square

Tech Policy Press - Technology and Democracy

Tech Policy Press is tracking laws and regulations, along with government investigations and litigation, that will shape the rules and accountability for tech companies.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe