On March 5, a bipartisan coalition of senators published a statement urging the Census Bureau and the Bureau of Labor Statistics (BLS) to improve their “collection, analysis, and dissemination of high-quality, timely data regarding artificial intelligence.” The question of whether — and how severely — AI will displace workers has become one of the most consequential and least settled questions in contemporary labor economics.  

 Accurate data is the backbone of our national response to labor displacement. To inform public policy, business decisions, and individual career choices, the BLS needs greater funding to generate accurate, AI-specific data using modern statistical methodology.  

 Over the past year, apocalyptic statements made by tech leaders have fueled widespread concern around AI’s future role in the labor force. Microsoft AI CEO Mustafa Suleyman, for instance, predicted in early 2026 that within 18 months, nearly all white-collar work would be automated. Studies from Stanford and Brookings predict significant disruption to white-collar employment. Meanwhile, reports from the Yale Budget Lab (YBL), the European Investment Bank (EIB), and Goldman Sachs assessed more moderate improvements in productivity.  

 For example, the EIB report found that firms that adopted AI increased labor productivity by 4%, with no adverse effects on their employment. The YBL report concluded that in the first 33 months since ChatGPT’s release in November 2022, there has been no major economic disruption in the labor market. By contrast, the Stanford study measured a 16% relative decline in employment among early-career workers aged 22 to 25 between 2022 and 2025. Additionally, it found that employment growth for young workers has been stagnant, with substantial declines in occupations heavily exposed to AI. In 2024, Brookings advocated for a working paper by the University of Pennsylvania and OpenAI with similar results, which estimated that roughly 19% of U.S. workers are in occupations where at least half of their tasks could be disrupted by generative AI.  

 The EIB used data from 12,000 firms across the EU and 800 firms in the United States. But in general, most of these studies rely heavily on data from the BLS, including Occupational Employment and Wage Statistics (OEWS), the Current Population Survey (CPS), and the Department of Labor’s O*NET task database. Key surveys for labor force statistics, like the CPS, are jointly sponsored by the Census Bureau and the BLS. 

 The BLS publishes 10-year occupational projections annually, but the agency itself acknowledges its projection methods are “not designed to capture extremely rapid technological change.” To enable the best possible decision-making by national stakeholders, better labor projections and data are both indispensable. 

 In November 2025, the Census Bureau added AI questions to the Business Trends and Outlook Survey (BTOS), a necessary starting point for the collection of AI-specific data. The BLS states clearly that the CPS needs modernization, as a response to both declining response rates and lack of AI-specific data. In April 2026, the BLS outlined what this modernization should look like:  

 The BLS has already begun the implementation of the internet self-response (ISR) instrument, which would run in conjunction with the CPS and provide critical additional data. The adoption of the internet survey response instrument and parallel surveys, as recommended in their report, addresses lower response rates and insufficient AI data. These modernization efforts go hand in hand with existing economic literature on AI data best practices. Common practices include the use of task-based AI exposure scores and LLM-based task annotations, which better capture the changing scope of AI’s impacts. Reform proposals tend to call for expanded data collection and better frameworks to process it. This proposal would cost an additional $60 million a year in funding. Currently, however, the agencies are unable to conduct testing at the scale necessary to ensure the accuracy of the new surveys due to funding constraints. 

 The big picture: The real (inflation-adjusted) funding for the BLS has declined by 13.8% since 2016. In January 2025, the BLS almost resorted to cutting the sample size of the CPS until a last-minute revision was made to the continuing resolution that allowed the BLS to spend CPS funds at a faster rate. The gravity of these potential cuts prompted two former BLS commissioners to urgently call upon Congress for funding.  

 For FY2026, the Trump administration’s budget request would be a $56 million decrease from the FY2025 budget, citing proposed reorganization efforts that would save money and move the BLS under the Department of Commerce. This sharp drop in necessary funding would almost guarantee furloughs and a stark decrease in the quantity and quality of BLS data at the moment it is needed most. 

 The BLS funding decline makes it difficult to maintain current operations, let alone respond to potential AI-induced labor volatility. For the BLS to generate concrete employment data in the world of AI, it will take funding to hire data scientists with AI expertise, to train existing staff on new tools and methodologies, and to secure reliable data-sharing agreements with private sector partners.  

 These reforms are essential. Without them, the labor market will remain impossible to measure accurately for the foreseeable future. Timely labor statistics are an invaluable asset when fast-paced economic conditions arise, so now more than ever it is necessary for the federal government to invest in its statistical agencies.

AI has embedded itself in America’s digital industries, such as software, finance, and entertainment. But for AI to raise productivity across the broader economy, efficiency and cost gains must extend to the physical industries that build homes, manufacture goods, and produce our food. Now, so-called “physical” AI is beginning to enable robots, machinery, and self-driving cars to perceive, reason, and perform actions in the physical world.

That’s essential. Physical industries such as construction, manufacturing, and agriculture have suffered from years of lagging productivity growth, along with a deep and worsening skilled-labor shortage. The good news is that the growth of AI in these industries may be accelerated by a collapse in sensor costs and dramatic increases in the quality of AI physics simulations. Across this dispersed and uneven frontier, machines that can perceive, reason, and understand physics may soon bring AI into new markets.

As with all technology-driven productivity, small improvements often seem like individual stepping stones until a full path is revealed. No one can promise hockey-stick-style exponential growth, but should it arrive, the effects on the cost of living and our politics would be profound. This brief will explore the state of play for physical AI across three critical sectors, construction, manufacturing, and agriculture, and consider what lies ahead for each.

Construction

The construction sector in the United States has been plagued by falling productivity since its peak in the late 1960s. Over the past twenty years alone, construction output per hour has fallen by 8%, while output per hour in the overall nonfarm business sector has risen by 40%.

Not unrelated, falling productivity is a key reason why prices in the construction sector have risen far faster than the overall inflation rate. From 2005 to 2025, the construction price index rose by 117%, double the 58% price increase for the economy as a whole.

Amid a lack of R&D in construction equipment, stringent land-use regulations, and a dearth of economies of scale, the construction industry’s plight has played its part in fueling the cost-of-living crisis. In simple terms, America would’ve gotten more bang for its construction buck, meaning more homes, factories, hospitals, and infrastructure from the same spending.

As a result, the output of the construction industry today is lower than in 2005 in real terms. Had construction prices risen at the same rate as the prices of the overall economy, the same nominal spending would have corresponded to a real construction output 30% above its 2005 level.

Physical AI could increase workers’ output through autonomous machinery. Among the many pioneers in the field, Bedrock Robotics has begun retrofitting excavators and bulldozers with lidar (Light Detection and Ranging) sensors, GPS, and cameras. Their package, titled the Bedrock Operator, completed the industry’s largest supervised-autonomous excavation on a 130-acre site. The start-up is targeting fully operator-less excavators in 2026 after its $270M Series B fundraiser, with total funding above $350M. Additionally, Built Robotics’ solar panel pile driver automates utility-scale solar pile driving: repetitive, precise, dangerous work where autonomy wins on cost and safety.

Caterpillar unveiled a generation of autonomous excavators, dozers, haul trucks, and compactors at CES 2026. This next generation of heavy equipment can work faster, longer, and more safely than human-operated equipment. WLTR, Wall Laying Terra-Based Robot, the brick-laying bot from GreenBuild, lays 108 square feet of brick per hour, roughly the output of five bricklayers, and requires the supervision of a worker from a safe distance. The robot uses glue instead of the carbon-intensive cement, which accounts for 7-8% of the world’s carbon emissions. Fastbrick Robotics (FBR) is pioneering similar bricklaying robotics, with successful Wall as a Service (WaaS) projects completed in Florida via their advanced construction robot, Hadrian. Hadrian is capable of building the load-bearing walls of a house, both external and internal, up to 360 blocks per hour, in less than a day.

By adopting these technologies, the construction industry can improve safety and output. To operate these novel technologies, workers must be upskilled and shifted towards higher-productivity, safer supervisory roles. Construction workers could become the managers of an AI-enhanced fleet of heavy equipment, shielded from environmental hazards and chronic injuries.

Manufacturing

U.S. manufacturing productivity, as measured by the Bureau of Labor Statistics, has been basically flat for the past 15 years. The issues in manufacturing are multifaceted, including geographic disparities, wage stagnation, competitive threats from China, and tariffs. Physical AI offers a chance to rapidly expand U.S. manufacturing capacity, develop more competitive products, and create stable, high-paying jobs.

First Solar, a solar manufacturer, has embraced physical AI across its production process. By utilizing AI computer vision in its newest Louisiana factory, deep learning tools spot defects and guide technicians to make real-time adjustments. The factory has a fully integrated production process and directly employs over 825 people. Average compensation sits at around $90,000, close to triple the area’s per capita income. First Solar is expected to produce more than 14 GW of solar capacity in 2026, with more than 17 GW of capacity when all five of its facilities are operational.

ABB Robotics and NVIDIA recently launched the integrated NVIDIA Omniverse simulation into RobotStudio (“HyperReality”), a physics simulation model for robotics, claiming to reduce costs by up to 40% while also accelerating time-to-market by 50%. Foxconn plans to pilot this in electronics assembly; the release to ABB’s 60,000 customers is scheduled for the second half of 2026.

Automaker giant Stellantis, NVIDIA, and Accenture are deploying AI digital twins across global plants. Digital twins pair physical production with an AI physics simulation of production to create a “predictive and autonomous” model driven by real-time data. In June 2026, LG and NVIDIA announced an AI-factory partnership that embeds NVIDIA’s robotics and physics models (Isaac, Cosmos, and GR00T) into LG’s “PhysicalWorks” robotics platform to connect procurement through delivery in real time. These NVIDIA models connect real-world information and visual data, robotics platforms for inference, and physics simulations to drive efficiency and increase safety. AI-powered robotics manufacturer MiR shipped the MiR1200 Pallet Jack, an AI pallet handler trained on 1.2M images that autonomously identifies, lifts, and delivers pallets in live warehouses. As with construction, physical AI’s integration with American manufacturing has been plant-by-plant rather than sector-wide, but the tools for widespread adoption are now commercial.

Agriculture

U.S. agricultural total factor productivity rose by only 1% from 2013 to 2023, according to the latest data available from the Economic Research Service at the Department of Agriculture. By contrast, agricultural TFP rose by 17% in the preceding 10 years (TFP measures output relative to all major inputs, whereas labor productivity measures output only relative to labor input).

Within the sector, impediments to further productivity growth remain, particularly for soft-flesh fruits such as berries, avocados, and tomatoes, which require hand-picking. Immigrant fruit pickers have been the backbone of this workforce. Recent policies from the Trump administration regarding ICE raids have raised questions about the labor supply of agricultural workers. Agricultural Secretary Brooke Rollins announced changes to the H2A visa program, effectively cutting the wages of workers between $1 and $7 per hour, depending on the state. The net effect of these policy changes remains to be seen. Still, the effective wage cuts create disincentives to innovation and automation, while questions remain about whether the labor demand will be met.

Either way, physical AI presents an opportunity to increase productivity and lower costs, which can translate into lower grocery bills for Americans. Monarch Tractor sells an autonomous, driver-optional electric tractor for specialty crops. Agtonomy and Kubota unveiled their line of autonomous tractors at CES 2025. The Agrobot E-Series is an electric, autonomous strawberry harvester that uses up to 24 decentralized robotic arms, each equipped with real-time AI, advanced vision sensors, and onboard graphics processing units, to assess and gently pick fruit without damaging it. The Agrobot E-Series illustrates how integrating sensor-based safety systems and automated labor solutions can bolster domestic agricultural supply chains, mitigate chronic farm labor shortages, and enhance worker safety.

Additionally, Carbon Robotics’ LaserWeeder uses computer vision to kill weeds without chemicals. John Deere has also developed its own line of autonomous tractors, equipped with 16 cameras, advanced computer vision, and remote-control capabilities for farmers via a smartphone app. The company has also developed a retrofitting kit that converts the 8R and 9R series into fully autonomous machinery. Although current use cases are primarily tilling, the company has also expanded into autonomous orchard tractors and commercial landscaping.

Policy Framework

To boost adoption, physical AI needs a policy framework that expands access to technology and establishes regulatory clarity for producers and users. Key initiatives could include a refundable adoption credit for small- and medium-sized enterprises, specifically targeting robotics-as-a-service in housing and specialty-crop production, so adoption isn’t tied exclusively to ownership. Regulatory clarity should also be built into an autonomous-vehicle framework for job sites; for example, by building on the BUILD America 250 Act and directing OSHA and NIST to develop rigorous safety and interoperability standards. Finally, restoring federal R&D funding to pre-2025 levels and indexing it for inflation would help sustain innovation.

Physical AI has yet to show clear effects on productivity or labor markets. No one can be sure when, or if, we are headed for a hockey-stick moment. Similar quiet progress occurred in generative AI during the years prior to the ChatGPT moment. Across construction sites, factory floors, and farm fields, the pieces are already in play. If the stepping stones connect, physical AI could raise output and lower costs across America’s physical industries, passing savings along to Americans through more affordable food, housing, and manufactured goods. Policymakers would be wise to adopt a proactive framework that allows physical AI to flourish, rather than a reactive one.

Politicians of all political stripes can be tempted to invest government funds in enterprises that directly compete with the private sector. PPI believes that such spending is rarely a good idea, especially in the highly competitive area of broadband communications. Too often, well-intentioned municipalities blindly go down the road of believing that they can build (and maintain) a complex network and provide broadband to their residents at a fraction of the cost of existing private companies. Unfortunately, past experience shows that this road is littered with wasted public resources and unhappy consumers.

That’s why we took a skeptical look at a new report from New York City’s Public Advocate, Jumaane D. Williams — one of three citywide elected officials in America’s largest city. In the report, Williams uses a purported crisis in broadband availability and affordability to call for New York to build a multi-billion-dollar “high-speed, low-cost citywide municipal internet service akin to a public utility.”

But such a costly project would be a mistake for New York City and its new mayor, Zohran Mamdani. As we show below, a closer  examination of official data shows that virtually every location in the city is already served by multiple high-speed internet providers, with average prices falling or stable in a broadband marketplace that is more competitive than it has ever been. It’s hard to see what would be gained by the costly and ongoing expense of a duplicative government-owned network in the city when networks are already deployed, and prices are dropping. The city has much more important fiscal priorities than laying out billions of dollars for unneeded infrastructure that would take many years to build. Moreover, the transition to this municipal broadband network, as outlined in the report, would disrupt an existing digital equity program that is already (and successfully) providing broadband services to hundreds of thousands of low-income households in New York.

Let’s start with the data. Inexplicably, the Public Advocate report uses out-of-date affordability and availability data in its effort to justify an expensive citywide municipal broadband network and backbone. However, more recent data undercuts the Report’s justifications and conclusions. Specifically, the “2025 Report on the Availability, Reliability, and Cost of High-Speed Broadband Services in New York State,” issued by the New York State Public Service Commission (PSC), provides a more up-to-date picture of broadband affordability and availability in the city. Table 1 describes the availability of high-speed internet service in NYC, meaning at least 100 Mbps download and at least 10 Mbps upload. When taking into account recent deployment data, the first column shows nearly every location in the city is “served”, defined as at least two internet service providers with at least one provider offering high-speed internet service. 

The second column shows that virtually every location in the city has more than one high-speed internet provider using wired or fixed wireless. In other words, almost everyone in NYC has a choice of high-speed internet providers. This hardly matches the broadband availability crisis painted by the Public Advocate’s report.

The more recent data also show the price of broadband in NYC is roughly in line with the rest of New York state. The first column of Table 2 reports the average price of 100Mbps internet in the five boroughs and compares it to the median of all New York state counties. We see very little difference. Indeed, the average price of broadband is falling in 4 out of 5 boroughs, and effectively flat in the fifth. The average price of high-speed internet in the Bronx, the poorest borough, has gone down by 12% over the past two years. That’s an achievement which should be lauded!  When the data is considered,  there’s no sign of a market failure or crisis that would justify spending billions of taxpayer dollars to create another broadband network in the city.

The Public Advocate’s report fails to grapple with the real cost of building out a city-wide muni broadband network, in a city with 8.3 million residents, one of the most complex underground infrastructure systems in the world, and no municipal utility to piggyback on. The report acknowledges the fiber backbone will cost more than $2.1 billion, based on an outdated January 2020 estimate. But that estimate predates the massive post-pandemic inflation in construction costs. Moreover, no discussion of the city’s debt capacity or capital budget constraints appears.

The Public Advocate’s report holds up Chattanooga’s muni broadband as a replicable model, but the comparison is deeply misleading. Chattanooga is a mid-sized city with a pre-existing municipal electric utility, suburban density, and a far simpler regulatory environment than New York City. 

In addition, the report wants to abandon “Big Apple Connect” — a program provided by Spectrum and Optimum and funded by the City to provide free internet to about 300,000 qualified residents in New York City Housing Authority buildings at a modest cost to the city government.  There is no indication that this program fails to deliver; yet, the report wants to close it down. NYC Mesh, the report’s preferred alternative to Big Apple Connect, is a volunteer-run organization with only 2,500 members after more than a decade of operation. It’s hard to imagine how to scale that up.

Too often, well-intentioned state (and federal) policymakers jeopardize a good program or project with a self-destructive devotion to another countervailing objective. One of the biggest disappointments of the Biden Administration was the BEAD program, enacted in 2021 to provide $42 billion in funding to fill in gaps in broadband coverage across the country, and especially rural regions. But when Biden left office at the beginning of 2025, not a single home had been wired with BEAD funds, in large part because well-meaning regulators loaded extra requirements onto the program. 

The New York City Public Advocate’s report would lead the citizens of New York City down the same sort of problematic path. Building an unnecessary municipal broadband system in the most crowded city in the country is a recipe for an expensive financial (and political) disaster. 

The broadband marketplace is intensely competitive. Cable is taking wireless market share, wireless is taking cable market share, satellite is nipping at the heels of everyone, and consumers are the winners. Broadband pricing has stayed well below the inflationary jumps that have been seen in many other industries/services.

This obvious competition, paired with low inflation, makes a new report from the Public Advocates Office (PAO) of the California Public Utilities Commission (CPUC) all the more curious. The report, “Broadband Competition and Pricing Strategies in California’s Urban Markets,” was based on an interesting new data set that looked at promotional broadband rates in four CA cities by detailed location, and compared those prices to the number of broadband providers and household income at those locations. The report’s goal, using regression analysis, was to show that fewer gigabit providers lead to higher prices and that providers were engaged in digital discrimination. 

We applaud the effort to assemble the data set. However, the analysis makes several fundamental mistakes. First, the pricing analysis left out important variables like population density (which could have easily been added at the census block level). A high-density area is generally cheaper to connect, on a per-household basis. As a result, high-density areas are more likely to attract new providers,  as well as leading existing providers to offer lower promotional rates. Conversely, low-density areas will typically have fewer providers and higher promotional rates. 

Thus, by leaving out density from their analysis, the report potentially found spurious correlations between fewer providers at a location and higher promotional rates. To put it another way, the first sentence of the executive summary says, “Broadband prices in California’s urban markets vary widely depending on the level and type of competition available to households.” But that’s tautologically true because competition was basically the only independent variable in their regression equation (the one exception was income, which we’ll discuss below). And disturbingly, the report even omitted results that did not show the desired correlation between price and competition. The report appendix notes that “Comcast is excluded from the regression analysis because its pricing strategy reflects large, market-wide discounts followed by secondary geographic variation that do not correspond to local competition intensity.” In other words, the Comcast regression did not show the desired results, so the report did not show it. In addition, the report acknowledges (note 37) that its regression analysis produces inconsistent results for Charter.

The second elementary mistake, related to the first, was the repeated confusion between correlation and causation. For example, the report asserts confidently that “San Diego has the most limited competition, with many neighborhoods served by a monopoly gigabit provider.” But as the table below shows, San Diego has half the population density of the other three cities, as well as being hillier. So it might be more accurate to say that out of the four cities, San Diego is the costliest, on a per-household basis, to lay fiber and cable, leading to fewer providers and higher promotional prices. Causality is very different than correlation.

Data: Census Bureau

Third, the report ignores intense competition in sub-gig markets, which many subscribers are intentionally choosing. As a result, the report only focuses on the four top fixed broadband providers, even though satellite providers, such as Starlink, and 5G internet providers, such as T-Mobile and Verizon, are available in many locations covered by the report. 

Finally, we come to the impact of income, which is the one non-competition variable in the report’s analysis. On its website, the PAO alleges digital discrimination in the California broadband space. However, the analysis in this report shows “providers do not systematically adjust promotional pricing based on income levels” (p 16). Truthfully, the report could have been entitled “No Digital Discrimination Found in California’s Urban Broadband Markets.”

The CPUC should be careful not to rely on this flawed study to make any policy judgments related to broadband. Rather, we should take comfort that the increasingly competitive marketplace for broadband is benefiting consumers. 

Introduction

Late last year, Japan joined the recent wave of countries attempting to regulate smartphone platforms such as iOS or Android when its new Mobile Software Competition Act (MSCA) went into effect. Like similar statutes around the world, the new legislation aims to give consumers more choices when it comes to how and where they purchase apps, while improving access for third-party developers.     

But compared to some of its international peers, Japan has pursued a more smartly tailored regulatory approach that should be seen as a model. It stands in especially stark contrast to the European Union’s overbroad Digital Markets Act (DMA), avoiding many of the safety and security issues that have been created by the EU’s effort. Here are four key areas where the Japanese and European approaches differ, underscoring how the MSCA offers a better approach. 

Security and User Choice

How to give consumers more choice about where they purchase apps without compromising their security is one of the most important questions in mobile platform regulation. Europe and Japan have taken vastly different approaches to the issue.

Under Europe’s DMA, platforms must allow users to download apps directly from third-party websites, without intermediaries providing protections or checks on in-app content. This approach creates substantial security risks, allowing for the distribution of malware or other content that skirts existing security reviews conducted by platforms.

Under the law, platforms are also required to let developers use an alternative payment system — so, for instance, your favorite music streaming app could direct customers to its own payment system and entirely bypass Apple or Google’s, which might charge the companies a fee.  As a result, users might no longer be able to use trusted platform payment options and instead be forced to share their financial information with an unfamiliar firm. The tools and protections that users have come to know and trust (like refund and fraud monitoring) could be removed as a result of poorly designed attempts to increase user choice. 

By contrast, the MSCA takes an approach that protects security while still encouraging user choice. By permitting measures “ensuring cybersecurity for smartphone use,” mobile platforms can ensure that alternate app stores include security protections or block criminal content. Meanwhile, alternative payment systems must appear alongside platform payment systems, expanding user choice with flexibility rather than reducing it. 

The MSCA’s approach acknowledges that competition regulation should expand the choices users have, not eliminate existing ones or push users towards less secure alternatives. 

Protecting Kids

Policymakers around the world are grappling with a difficult question: how do we keep young people safe online in an increasingly digital world? There are no easy answers to this question. But at the very least, online competition rules shouldn’t make it more difficult to protect kids. 

Unfortunately, the DMA, passed in 2022, fails on this front. For starters, it does not include explicit protections for minors. As a result, children are treated the same as adults when it comes to the DMA’s rules on alternative app distribution and payment services, discussed above. Since the law forces platforms to allow relatively free access to sites and systems outside of the platform owner’s control, efforts to restrict the content that kids can access could be considered a violation of the DMA. For example, age restrictions for apps distributed outside the app store may not be permissible under the DMA; platforms have already been forced to allow apps containing explicit content to be installed through alternate app stores. Lacking a carveout for youth protections, kids could be left with unmitigated access to explicit or harmful content online.

Unlike the DMA, the MSCA permits measures “safeguarding youth who use smartphones.” Measures that might face legal challenges under the European approach can remain in place in Japan. Tools like limiting transaction links in apps designed for children, restricting access to alternative markets through controls at the operating system level, and limiting targeted advertising to minors are all possible through the MSCA, empowering parents. 

As PPI has previously explored, parents are strongly supportive of these sorts of controls, which enable them to make decisions about their kids’ online access. The MSCA’s approach does not resolve all of the difficult questions about kids’ online safety, but it provides the flexibility needed to maintain existing safeguards while still preserving competition. 

Interoperability and Privacy

Interoperability — allowing third parties to interface with a platform’s systems and data— is an admirable strategy for strengthening competition and helping users get more out of their devices. But it is not without tradeoffs: integration inherently requires control and access to potentially sensitive data. Competition regulation should be selective about where and how interoperability is mandated in order to maximize the benefits for users while maintaining safety as much as possible.

The European approach forces platforms to provide third parties with sweeping access to user data for interoperability purposes. For example, mobile platforms have faced requests to hand over the full contents of users’ notifications or the history of Wi-Fi networks they have connected to, regardless of how the third party intends to use the data. Notification contents could expose two-factor authentication codes or private details, while Wi-Fi history could reveal where and how a user is spending their time. With no option for platforms to deny requests for sensitive user information, third parties may maliciously harvest and monetize data for their own gain, all while consumers remain unaware of risks. These overly broad interoperability mandates harm user privacy and could eventually erode trust in platforms,  hurting the market for all developers.

Japan’s competition law takes a narrower approach to interoperability access. The MSCA requires that requests remain “proportionate to the competition related problems at hand” and allows platforms to reject inappropriate data access attempts. Platforms can also reject requests from parties who are legally required to share collected data with foreign governments, keeping user data safe. These measures mean that opportunities for interoperability, which benefit users, can remain in place, while those that exploit them are rejected.  

Innovation and Intellectual Property

Governments also need to balance their desire to encourage competition with the need to incentivize innovation. Requiring platforms to share features or access with competitors can provide users more choice, but it also weakens the returns from research and development. This can lead to stagnation as companies find themselves unwilling to invest in innovation. In designing competition regulation, policymakers are forced to make a choice about how to strike this balance, and the DMA and MSCA represent meaningfully different answers with real consequences for users.

The DMA’s “interoperability by design” approach means that platforms are often required to share features and IP with third parties without compensation, including early notice about coming updates. This gives competitors valuable insight into platforms’ future plans, and introduces significant costs for platforms to make new systems and features compatible while competitors’ development costs are effectively subsidized. 

These misaligned incentives mean that platforms may withhold or delay new features, leading to a worse experience for users. Recently, Apple has delayed iOS features like Live Translation or withheld others like iPhone Mirroring entirely in Europe as a result of the DMA. Apple argues that because of concerns over privacy and compliance with interoperability requirements, such delays are likely to continue. Today, European users have a limited product compared to their international counterparts, not due to technical limitations but because of the high costs of legislative compliance.

Japan’s “proportional interoperability” approach is narrower and contains protections for “legitimate exercise of intellectual property rights,” including the ability for platforms to charge for interoperability access. Platforms maintain the right to evaluate whether to implement interoperability access on some features, allowing them to more effectively use resources and preserving incentives for R&D investment. Tellingly, Japanese users have so far not faced the same feature delays or limitations that European users have. The results show that the MSCA’s proportional approach can still address genuine competition concerns without severely damaging incentives for innovation.

Conclusion

The DMA and MSCA present two significantly different approaches to increasing competition in the smartphone ecosystem. The DMA’s hardline, no-exceptions approach has lofty ambitions, but has already led to negative tradeoffs for consumers, including reduced protections for minors, privacy risks with interoperability, and delayed features. While the full impact of the still-young MSCA remains to be seen, its moderate approach appears poised to avoid the same pitfalls that have hampered the DMA to date. 

As both laws mature and their impacts are fully understood, the comparative outcomes will be instructive for policymakers around the world considering similar legislation. Effective competition policy should expand user choice without major sacrifices to security, privacy, or other protections that users value, and Japan’s efforts show this balance is achievable.

Congress is moving swiftly ahead on legislation that would require smartphone apps to verify the ages of their users in order to protect children’s safety online. But with full markups on several bills scheduled for the coming weeks, lawmakers face an important choice between competing approaches.

Among the leading bills, H.R. 6333, the bipartisan Parents Over Platforms Act (POPA) introduced by Reps. Jake Auchincloss (D-Mass.) and Erin Houchin (R-Ind.) stands in strong contrast to H.R. 3149, the App Store Accountability Act (ASAA) introduced by Rep. John James (R-Mich.). Though both bills aim to safeguard kids through requirements on app stores, POPA stands out as a more practical, privacy-forward, and parent-aligned approach. Here are four key policy areas where the two bills differ. 

Shared Responsibility

Checking an app users’ age is a complex task, and responsibility for it should fit with the actual roles of app developers and stores. While both bills enlist mobile app stores in the age assurance process, ASAA establishes onerous requirements that would require them to collect government IDs from all users — minors and adults alike — regardless of what kind of app they want to download.

While app stores may be well-positioned to use basic age information to limit the apps younger people can download, they have little control or knowledge of what happens once an app is installed. The ASAA would make the point of download the only major check on online safety, with app developers holding minimal responsibility for providing safer experiences after their product is loaded onto a child’s phone. This strategy could inadvertently lead to poorly applied restrictions with content minimally tailored to be age appropriate and little control of how apps are actually used.

By contrast, POPA proposes a shared responsibility across the ecosystem, requiring app stores to conduct age checks at the point of download and developers to do the same when consumers use certain parts of their apps. As PPI wrote last month, parents believe strongly that verification should go beyond a one-time check. 

Consent Fatigue

For parental consent to be meaningful, it must be sustainable: Too many requests lead users to accept terms without reading them, much the way most of us now automatically click through the ubiquitous cookie consent banners websites started displaying following enforcement of Europe’s GDPR. This phenomenon, known as consent fatigue, should be carefully considered in the design of an age verification framework.

POPA gives parents tools to manage kids’ access without becoming overly frequent or demanding. For example, parents are able to restrict access by category or age rating, rather than at every app download. Giving parents these tools means they still can still make meaningful decisions about their kids’ activity without being inundated by routine approvals that might cause them to tune out. 

Though designed with good intentions, ASAA’s approach is much more likely to overwhelm parents. The bill would require app stores to receive parental consent during each and every download request. Even if parents decide that their kids are ready to download some kinds of apps independently, ASAA does not provide a mechanism to let them do so. And with requirements to receive additional parental consent after “significant changes” are made to any app, requests are likely to be frequent.

Parental Control & Data Sharing

When handling sensitive personal information, privacy and choice should be foundational priorities. The age assurance process should strive to minimize data collection and sharing, obtaining only the information needed for age assurance and nothing beyond.  While POPA gives parents the agency to decide when and where their child’s age is shared, ASAA mandates broad sharing without consent. 

ASAA requires all new users to undergo age verification, and developers of all apps – even those without age restricted content — to receive information about all users’ ages by default. This approach violates the principle of data minimization and puts all users at risk. Even if a user wishes to download an app that does not include age-restricted content, like a notetaking app or their favorite coffee shop’s app, they will still be required to undergo the age verification process. Parents and other users have no choice over the collection and sharing of their information.

POPA takes a narrower approach, allowing users to declare their age and giving parents the ability to choose to share age information with developers. While the bill encourages app stores to use techniques like age estimation to provide age assurance, it does not require it. 

Legal Considerations

Crucially, the concerns over scope and applicability considered in this piece are not purely speculative. Last month, a federal judge temporarily blocked Texas’s state-level age verification law on the grounds that it was “exceedingly overbroad” and “unconstitutionally vague.” If Congress is serious about protecting children and giving parents choice, it should pursue a legally durable approach that can withstand the same first amendment challenges that the Texas law faced. POPA’s measured scope, with a defined set of covered applications and focus on parental consent and choice, appears up to this legal scrutiny. As markup approaches, lawmakers now have an opportunity to advance legally durable and practically designed age assurance legislation. Congress should choose POPA.