China's President Xi Jinping and President Donald Trump are looking towards the sky, wearing matching suits and ties, as Trump points his finger in the air. Military brass band members are seen in the foreground. 

President Donald Trump and China's President Xi Jinping watch a flyover during a military review in the Rose Garden of the White House, Thursday, Sept. 24, 2026, in Washington. (AP Photo/Jacquelyn Martin)

US Leadership on Open-Weight Models is Key to Global AI Safety 

Immigration and Customs Enforcement (ICE) agents face off with protesters during a shift change outside of Delaney Hall Immigration Detention Center on June 7, 2026 in Newark, New Jersey. (Photo by Adam Gray/Getty Images)

Black Migrants Bear Brunt of ICE's Facial Recognition Push

A group of teenagers hold their cellphones in a circle, Yorkshire, United Kingdom. (Edwin Remsberg / VWPics via AP Images)

Where Does the KIDS Act Fit in the EU’s Digital Rulebook?

European Commissioner for Tech Sovereignty, Security and Democracy Henna Virkkunen. (AP Photo/Marius Burgelman)

Europe’s Digital Sovereignty Has a Funding Gap

A view of the Berlaymont European Commission building in Brussels, Belgium on July 20, 2026. (Photo by Jakub Porzycki/NurPhoto via AP)

The Overlooked Issue Undercutting EU’s Out-of-Court Dispute System

The silohuette of two children seen from below who are looking down through the black mesh fabric of a trampoline with the sun shining behind them.

A photo from 2008 of children playing on a trampoline at the Soweto Hospice in South Africa, which was funded by the US President’s Emergency Plan for AIDS relief, PEPFAR. (AP Photo/Denis Farrel)

The Life and Death Stakes for Africa of a World That Doesn't Agree on the Truth

UN Secretary General Antonio Guterres holds a press conference at the UN headquarters in New York on Sept. 16, 2026, at which he spoke “about rising concerns regarding the advance of artificial intelligence.” (Kyodo via AP Images)

UN Secretary General Antonio Guterres holds a 

 at the UN headquarters in New York on Sept. 16, 2026, at which he spoke “about rising concerns regarding the advance of artificial intelligence.” (Kyodo via AP Images)

Who Is AI Risk For? Centering People in the AI Risk Conversation

Trendy Halftone Collage Hand and stock arrow move up. Market statistics chart. Finance business data graph. Investment growth. Trending upwards. Contemporary vector illustration art

We Need Google Trends for AI

A Flock brand automatic license plate reader camera (ALPR) is attached to a utility pole along 16th Street Northwest on August 19, 2026 in Washington, DC. (Photo by Chip Somodevilla/Getty Images)

A Flock of Problems with Mass Surveillance

European Commission President Ursula von der Leyen and Executive Vice President Henna Virkkunen speak at a press conference presenting the EU KIDS Act in Strasbourg, France, Sept. 17, 2026. Source: von der Leyen/X.

How the EU Kids Act Can Promote  Interoperability for Online Safety

A green and yellow three-wheeler on a street moves past BRICS summit banners bearing images of Indian Prime Minister Narendra Modi, Russian President Vladimir Putin and Chinese President Xi Jinping

BRICS 2026 summit banners bearing images of Indian Prime Minister Narendra Modi, Russian President Vladimir Putin and Chinese President Xi Jinping in New Delhi, India, Sept. 12, 2026. (AP Photo/Mukhtar Khan)

Before India Exports Digital Public Infrastructure, It Should Define It at Home

CHONGQING, CHINA - AUGUST 10: In this photo illustration, the logo of Meta is displayed on a smartphone screen on August 10, 2026 in Chongqing, China. (Photo by VCG/VCG via AP )

Meta Brings Community Notes to LatAm Without Showing What Success Looks Like

The Léon Thévenin near the Buitenratel, Belgium in 2009. Hans Hillewaert/Wikimedia Commons/CC by 4.0

The Léon Thévenin near the Buitenratel, Belgium in 2009. Hans Hillewaert/

The UN’s AI Agenda Must Prioritize Undersea Cables

US President Donald Trump and Chinese President Xi Jinping attend a welcome ceremony at the Great Hall of the People on May 14, 2026 in Beijing, China. (Photo by Alex Wong/Getty Images)

Trump and Xi Should Talk About Worms

A military band conductor with raised arms with a view from below of an ornate ceiling with a red star in the middle

A military band conductor leads during a ceremony to mark the 105th anniversary of the Chinese Communist Party at the Great Hall of the People in Beijing, July 1, 2026. (AP Photo/Ng Han Guan)

US, China Psy-Op Accusations Signal Deep Distrust on AI Regulation

Secretary-General Antonio Guterres speaks during press conference ahead of General Assembly High-level Week 2026 at UN Headquarters in New York, NY on September 16, 2026.(Sipa via AP Images)

Can the United Nations Help Slow the AI Race?

The setting sun reflects off an Anduril Sentry tower in Luna County, New Mexico. EFF/Wikimedia Commons October 2022

The setting sun reflects off an Anduril Sentry tower in Luna County, New Mexico. EFF/

The Influence of AI on Human Decisions in DHS Surveillance

Interior view of House of Commons at Canada Parliament. Ottawa, Canada - December 15, 2025. Shutterstock

Interior view of House of Commons at Canada Parliament. Ottawa, Canada - December 15, 2025. 

Canada Is Not Banning Teenagers from Social Media — It’s Trying Something New

ICE agents stand guard outside Delaney Hall detention center in Newark, New Jersey, on Thursday, May 28, 2026. (Photo by Cristina Matuozzi/Sipa via AP Images)

From Records to Raids: The Questionable Legality of ICE’s Palantir ELITE System

European Commission President Ursula von der Leyen delivers a speech at the annual State of the European Union at the European Parliament in Strasbourg, Wednesday, Sept 16, 2026. CC-BY-4.0: © European Union 2026– Source: EP

Europe Says Its AI Rules Are Enough. AI Agents Are Testing That Claim

Earlier this year, thousands of United States Immigration and Customs Enforcement (ICE) and Customs and Border Protection (CBP) agents were deployed across the Twin Cities and Minnesota in a massive federal immigration enforcement campaign launched by the Department of Homeland Security (DHS). The campaign, Operation Metro Surge, extended beyond borders and brought immigration enforcement directly into Minnesota’s communities, inciting widespread terror among communities and their families. According to 

, community members were so terrorized by ICE and CBP agents' excessive and unpredictable use of force, they restricted their movement and sheltered in their homes for weeks.

This siege disrupted nearly every aspect of daily life in Minneapolis, threatening residents’ ability to work, attend school, access health care, and seek legal protection. The City of Minneapolis 

 that the campaign resulted in 4,000 arrests and an estimated $700 million impact on the community and City government operations. Facial recognition technology played an integral role throughout this campaign, as federal immigration agents used 

, a smartphone facial-recognition tool that allows agents to scan the face of individuals they encounter and search it against multiple government databases. The use of this technology in this campaign reflects how facial recognition can be used to expand ICE’s reach, transforming immigration enforcement into a broader surveillance dragnet capable of reaching immigrants and citizens alike.

The expanded use of this technology is particularly concerning for Black migrants and other migrants of color, since facial recognition performs differently across racial groups. Black migrants are disproportionately exposed to the surveillance systems using facial recognition in the first place. Drawing on a 

 by the Black Alliance for Just Immigration (BAJI) and the Immigrant & Racial Justice Solidarity Clinic and International Justice Clinic at UC Irvine School of Law to the UN Special Rapporteur on contemporary forms of racism, this article examines how facial recognition has become embedded throughout US immigration enforcement, including ports of entry, ICE enforcement, and biometric databases; how its deployment reproduces and amplifies existing racial inequalities; and why that expansion raises concerns under international human rights law, particularly the International Convention on the Elimination of All Forms of Racial Discrimination (ICERD).

The use of facial recognition during Operation Metro Surge is not unique nor isolated; it is one example of a much larger system in which the government uses biometric technologies throughout immigration enforcement. Facial recognition is now used well beyond identity verification at airports and ports of entry. The government now uses it across immigration enforcement; from ICE agents scanning faces during street-level enforcement operations, to remote check-ins through apps like 

, to large-scale DHS databases that store biometric information. These uses of facial recognition do not necessarily operate separately. Biometric information collected in one encounter can be retained, combined with other personal information, and accessed or shared across government agencies, allowing information collected for one purpose to later be used in other immigration and enforcement contexts. This system continues to grow. In July 2025, a 

 toward CBP technology, including expanded border surveillance, accelerating the federal government’s investment in the technologies that make this system possible.

This expanded use of facial recognition is particularly significant because there are racial disparities in both its design and deployment. facial recognition has 

 for darker-skinned faces since it is trained on white data and relies on databases which are more likely to categorize darker skinned people as “risks” or “threats” because of increased enforcement and surveillance of black and Latinx neighborhoods in the US. These existing patterns of racialized surveillance persist in the deployment of facial recognition, since they continue to be disproportionately exposed to surveillance technologies that perform less reliably on them. As a result, expanded use of facial recognition can subject Black migrants and other migrants of color to heightened surveillance, misidentification, wrongful stops or detention, and the collection and retention of their biometric information for further use across multiple enforcement contexts.

Further, as seen during Metro Surge, these harms can extend beyond the individual migrant to their families and communities. During the operation, US citizens were recorded. As federal agents were deployed into Minneapolis and the Twin Cities, they would 

 residents, target individuals at protests or those observing enforcement activity, or go into public spaces they assumed to be immigrant spaces. When federal agents first encounter an individual in these operations, they do not immediately know their status, so they collect identifying information about them regardless of immigration or citizenship status. This expands immigration surveillance beyond those ICE initially sought. As a result, other people in the same spaces, such as family members and neighbors, can also be subjected to this surveillance. Therefore, facial recognition does not simply reflect existing racial inequalities; its use can reproduce and amplify them.

These racialized harms also implicate the US’s binding obligations under international law, which provides a framework for evaluating both the technologies themselves and state policies that authorize and expand their use. Under the International Convention on the Elimination of All Forms of Racial Discrimination (ICERD), the United States has a binding obligation to (1) prevent both direct and indirect racial discrimination, (2) refrain from engaging in and prevent acts of racial discrimination, and (3) mitigate structural racism by amending policies that perpetuate it. ICERD also requires states to regulate third-party vendors, ensure equal treatment before the law, and guarantee an effective remedy for people subjected to racial discrimination. Applied to facial recognition, this convention requires that the US look beyond whether the technology intends to discriminate along racial lines and examine how the government’s design and deployment of facial recognition impacts Black migrants and other migrants of colors in practice.

Racial bias is a design element of facial recognition

The racial bias inherent to facial recognition technology’s function starts before it is ever used at a border. Facial recognition systems are trained on datasets that are more than 80 percent light-skinned. Since there is not equal representation in the data the technology learns from as it develops, it becomes more accurate when it is deployed against those with lighter-skinned tones than darker skin tones. 

Further, the cameras used to capture new images are typically calibrated for lighter skin tones, so image quality for darker faces is degraded before an algorithm processes them. As a result of these design choices, the 

 for facial recognition is as low as 1 percent for lighter-skinned men and as high as 34.7 percent for darker-skinned women. So, facial recognition performs differently depending on race and gender. In the case of one private vendor, errors were attributable to darker-skinned subjects 93.6 percent of the time. For another vendor, the errors were attributable to women 95.9 percent of the time. Further, facial recognition systems compare a person’s face against large databases of stored images. 

Since Black communities are policed at higher rates, they are also 

 in those databases. Therefore, the databases that facial recognition draws from are already skewed against Black people as existing patterns of racialized surveillance are embedded within its design.

Surveillance at the border and ports of entry

Today, nearly every point of entry into the US has become a biometric checkpoint. Through its 

 CBP now uses facial recognition at all international airports on entry, at 66 airports on exit, across 39 seaports, and at multiple land crossings. 

The government’s use of biometric surveillance at such ports is only poised to expand. A proposed DHS rule would allow USCIS to require non-citizens of all ages to routinely submit biometric information throughout the immigration process until they become US citizens. This proposed policy would also expand the types of biometric information USCIS can collect, along with the circumstances in which this information can be reused.

Facial recognition does not stop at the border; it follows migrants into the interior of the United States. After entering the US, some migrants are placed in ICE's Intensive Supervision Appearance Program (ISAP) instead of being held in detention. Most recently, it has been 

 that the program consists of around 200,000 people. As part of the program, most participants are required to use the SmartLINK app, which uses facial recognition to verify migrants’ identities during ICE check-ins and confirms their location during the check-in. This app is used by 

 of the program’s participants. While ICE claims that location is only tracked and recorded during required check-ins, 

 by The Markup and Documented found that, for two SmartLINK users whose devices were examined, the app accessed location information when it was launched, as well as during required check-ins.

Glitches and malfunctions within the app can have severe consequences for migrants’ immigration processes and possibilities of re-detention: one user missed one check in because his phone died, and another because the app malfunctioned. Technological malfunctions like this can potentially lead to findings of non-compliance, that may ultimately result in increased reporting requirements, re-detention, and separation from family and community. Users are aware of these risks. Some users state that they feel “constantly exhausted and anxious” living under this app’s surveillance. In one law review account, a woman placed in the program—a composite of the author's clients—

 that she was scared that, should the app malfunction, she would be re-detained and separated from her siblings.

The facial recognition systems discussed above, such as CBP Home, airport scans, and SmartLINK, do not operate independently. They all feed into a larger DHS system for collecting, storing, and sharing biometric information. IDENT, DHS’s current biometric database, is 

 by the Homeland Advanced Recognition Technology System (HART). Instead of only storing fingerprints and faceprints, HART goes further, connecting those biometrics with other information, including political affiliation, religion, location, and known associations. The database is not just for CBP; it can be accessed by the Department of Defense, the Department of Justice, state and local police, and foreign governments.

The lack of transparency surrounding this database has led watchdogs to 

 a “black box,” since people are not notified before their information is collected, and they don’t know what information is stored within the database and who has access to it.

The expansion of facial recognition throughout immigration enforcement is also largely driven by private corporations, who develop and operate key parts of this surveillance structure. Clearview AI, which scrapes photos from social media and the open web, holds a 

 with Homeland Security Investigations (HSI) 

 for and was founded by a figure who has documented associations with white-nationalist and far-right figures. Palantir Technologies builds the data-integration infrastructure that lets ICE fuse biometric, location, and biographic data into a single profile. These companies consider key information about how their proprietary technology 

 as confidential business information that competitors should not be able to access, so they argue that they are 

 the systems relied on to produce particular results. This makes it very difficult to challenge or independently evaluate these systems, limiting the companies’ accountability.

Further, data-sharing extends beyond the US border. At the time our 

 was submitted to UN Special Rapporteur, the government had biometric-sharing agreements or letters of intent with five countries: 

. Since then, the government has expanded biometric data sharing throughout the region. Most recently, in July 2026, DHS 

 with the Caribbean Community’s Implementation Agency for Crime and Security (CARICOM IMPACS), creating a framework for sharing biometric and biographical information across the Caribbean. The agreement includes particular cooperation with Caribbean countries that operate 

 programs, including Antigua and Barbuda, Dominica, Grenada, Saint Kitts and Nevis, Saint Lucia, and Saint Vincent and the Grenadines. This means that US immigration enforcement no longer begins when someone physically reaches the US border. Instead, because biometric information is shared with other governments, US authorities may identify and monitor migrants while they are still traveling through Latin America, exposing migrants to profiling and refoulement long before they reach US soil.

From historic Black Codes to facial recognition Black codes

The racialized, uneven consequences of these technologies often go unscrutinized. This, in part, is because the government justifies facial recognition by arguing that it is preventative rather than punitive. More specifically, it argues that facial recognition is necessary for national security, prevention of fraud, management of borders, and minimization of crime. However, in doing so, these technologies make determinations about who is considered as a “risk” or “threat.” 

These determinations are not neutral; they often reflect older racialized assumptions about riskiness that predate facial recognition by centuries: lantern laws, which required Black, Indigenous, and mixed-race enslaved people in colonial New York to carry lanterns after dark so they could be more easily monitored; slave patrols, which monitored and controlled enslaved Black people; Black Codes, which criminalized Black life after emancipation; and COINTELPRO, which surveilled Black civil rights and liberation movements. In this reading, facial recognition does not simply police the border; it helps construct it, transforming it into a network of surveillance that digitizes colonial logic and sorts bodies into categories of belonging and threat along racial lines.

In their report to the United Nations Special Rapporteur, BAJI and the UCI Clinics propose that the solution to decolonizing this technology is embedding a collectivist rather than individualistic worldview into how it is designed and governed. An individualistic approach asks how a system may efficiently identify an individual. On the other hand, a collectivist approach asks how a system may affect relationships, communities, and human dignity.

2001 Durban Declaration and Programme of Action

, which identified colonialism as one of the historical roots of contemporary racism, the report argues that addressing discrimination in facial recognition technology requires confronting the historical structures that continue to shape its design, deployment, and use. This approach mirrors Jose Cossa’s contribution, 

, an approach derived from the African philosophy of Ubuntu, which is rooted in the idea that “a person is a person through persons.” In other words, Ubuntu understands personhood as fundamentally relational rather than individualistic. Applied to technology governance, that means evaluating systems not only by whether they function efficiently, but also by how they affect human dignity, communities, and relationships.

This decolonial praxis offers an alternative to the individualistic assumptions that often shape Western approaches to technology design, asking developers and agencies to examine not just what a system does, but why it was built that way, and to give the communities most affected by it, including the African Diaspora, a meaningful role in its future design and oversight.

The consequences of expanding immigration surveillance are already visible, as facial recognition technology has become an additional tool of oppression employed against our country’s most vulnerable communities. In Springfield, Ohio, in the aftermath of the 

 of hundreds of thousands of Haitian immigrants’ Temporary Protected Status (TPS), ICE agents 

 at least thirty local Haitians at the Butler County Jail. Haitian immigrants in Springfield have been shackled with ankle monitors, have had their biometric information collected through facial recognition technology without their consent, have been zip-tied in front of their children, pulled off of their job sites and out of their homes. They are now scared to leave their homes, go to grocery stores, or take their children to school. Through these methods of intimidation and oppression, it is evident how facial recognition technology has expanded the scope of the subjugation Black and Brown people can experience at borders, in their homes and communities, and as they move through the world.

While the racialized error rates associated with facial recognition technology are indeed concerning, facial recognition technology still causes real harm when it operates as intended, as it functions within existing systems of power. In these systems, it furthers existing patterns of over-policing communities of color that are disproportionately subjected to immigration enforcement. Therefore, the question is not simply whether the technology is accurate. Rather, we must examine who is subject to the increased surveillance it offers and what happens with the information it collects.

From pouring in billions of euros of public funds to build 

, Brussels wants to reduce its dependence on Washington and position the European Union as a legitimate force in the global Digital Great Game.

Yet just as EU officials flex their muscles — including the 

 in the bloc’s next seven-year budget for cloud computing, semiconductors and other strategic technologies — Europe has left itself exposed to outside forces.

Brussels’ growing stable of digital rules, including the recently announced 

, relies heavily on outside groups to provide the 

live fact-checking and election monitoring

defense of citizens’ fundamental rights

These groups, collectively, are Europe’s Democracy Stack.

They are the independent media outlets, disinformation research groups, regulatory compliance organizations and internet governance bodies that represent the strategic public infrastructure at the heart of the bloc’s digital and geopolitical ambitions.

At their core, they provide the democratic accountability layer to the EU’s digital sovereignty agenda. These groups hold both companies and governments to account in an increasingly digital age. They also connect European citizens to the bloc’s long-term plans to become more sovereign and more autonomous.

Yet Europe’s Democracy Stack is anything but sovereign.

 from the Atlantic Council’s Democracy + Tech Initiative and the Global Disinformation Index of almost 100 organizations at the center of this public-interest infrastructure quantified, for the first time, how much Europe relies on others — and the United States, in particular — to pay for the bloc’s digital aspirations.

Currently, only 12 percent of the annual budget for Europe’s Democracy Stack is funded by EU institutional funds, based on a review of groups’ statutory filings, the annual reports of philanthropic organizations and a tallying of EU and non-EU government grants.

In contrast, almost 60 percent of the capability’s annual 144 million euro operating budget for 2024, the latest full-year figures available, comes from either the US government or US philanthropic organizations. The remainder came from EU national governments or commercial and membership income.

That includes non-EU institutional funds to protect Europe’s elections from foreign interference; fund strategic litigation to defend the bloc’s fundamental rights; and support the regulatory compliance functions of groups that both the European Commission and EU member state governments routinely rely on to enforce Europe’s digital rulebook.

Such EU dependency on US funding did not appear overnight.

Until recently, Europe’s digital rights, election integrity and democratic resilience agenda went hand in glove with the priorities — and deep pockets — of both Washington and major US philanthropic organizations. This alignment allowed Brussels to 

 for Europe’s Democracy Stack with that of like-minded non-EU government agencies and foundations, which received a global halo effect of backing Europe’s 

Between 2021 and 2024, it helped to expand the collective funding for Europe’s Democracy Stack by 51 percent, cumulatively, at a time when Brussels was writing, implementing and enforcing the Digital Services Act, Digital Markets Act and Artificial Intelligence Act.

This expansion, however, was built on faulty ground. Between 2023 and 2024, for instance, almost 40 percent of groups reported a decline in their income, according to regulatory filings. In the wake of a changing of the guard in the White House, US government funding has dried up.

Washington has pivoted hard away from such digital rights work, and now even 

 that openly challenge aspects of Europe’s Democracy Stack. US philanthropic organizations have similarly cut back on supporting European groups. They 

 toward domestic concerns or doubled down on investments targeting artificial intelligence.

Confronted with this financial shortfall for groups central to the bloc’s digital sovereignty project, EU officials have failed to respond.

What current public funding exists for Europe’s Democracy Stack is overly bureaucratic, short-term in nature, and misaligned to the bloc’s digital needs, based on a 

 from the Atlantic Council’s Democracy + Tech Initiative that drew on a series of EU member state workshops.

In part, this failure is political. Just as US federal officials 

 Europe’s digital-focused groups for censoring Americans’ free speech, so too have some EU politicians, including those in Germany, 

 that Europe’s scarce public funding should not be earmarked for civil society groups.

At a time of sluggish European economic growth and belt-tightening 

 in Europe’s next seven-year budget, such reticence is understandable when many lawmakers believe supporting these outside groups is a luxury, not a necessity.

Yet this is a fundamental misreading of Europe’s Democracy Stack.

As part of the EU’s next budget cycle, officials have determined that reducing the bloc’s structural dependencies on other countries, most notably the US, is a political imperative. That includes a so-called 

 worth roughly 400 billion euros to fund research and deployment of strategic technologies, among other priorities.

Europe must apply that same logic — that reliance on outsiders for the bloc’s digital future is no longer acceptable — to the public-interest capabilities necessary to provide democratic accountability to the public money spent on artificial intelligence, semiconductors and the implementation of the EU’s digital sovereignty agenda.

Reducing that dependency is the missing layer of Brussels’ political ambitions. It can only be resolved through greater EU institutional support, via the next bloc-wide seven-year budget, for the public-interest infrastructure that defines Europe’s Democracy Stack.

That is where Brussels needs to act, and the fix isn’t complicated.

Currently, Europe only provides roughly 17 million euros a year for the fact-checkers, election-monitoring watchdogs and digital rights groups upon whose work the bloc’s digital rulebook — and global reputation — depends, based on the analysis of the finances of Europe’s Democracy Stack.

Matching what the US government and philanthropies have traditionally paid would mean Brussels providing closer to 85 million euros a year. Importantly, it must be committed for several years at a stretch instead of doled out grant by grant to ensure Europe’s Democracy Stack can be the public infrastructure it was designed to be.

Next to the 400 billion euros which Brussels wants to fork out for high-end semiconductors and AI infrastructure, such figures are mere rounding errors. When confronted with threats to its sovereignty, Europe found the money for glitzy AI gigafactories. It now also needs to find the money for its Democracy Stack that keeps its digital rules accountable to citizens.



As President Trump opens the White House doors to Chinese President Xi for a high-level 

 beginning today, AI safety is top of mind for many 

as we’ve heard all month. But there is a critical question that has received surprisingly little attention: what should the US government do about ever-advancing open-weight models?

Open-weight models are models whose weights, or numerical parameters, are released publicly. That means anyone can download and modify the models. Some open-weight models require specialized hardware to host, while others are small enough to run on a 

—including GLM 5.3, Kimi K3, DeepSeek V4.1-Flash, and Qwen3.8—are Chinese. Open-weight models generally lag the best American closed-weight models, but the gap 

. The UK government’s AI Security Institute 

 in July that leading open-weight models’ cyber capabilities are only four to seven months behind those of the best closed models.

From an AI safety perspective, open-weight models present unique challenges. Because anyone can download open-weight models, users can often 

 the guardrails that are meant to prevent people from using AI to hack others or build a bioweapon. And once a model has been downloaded, the developer cannot recall it, meaning that there is no “kill switch.” 

 that these features make open-weight models riskier. It might therefore be tempting for the US to prohibit open-weight models and cede the global open-weight market to China. Indeed, 

 predict that open-weight models will be banned “before too long.” But that would be a big mistake. Instead, the US should make it a national priority to develop safer open-weight models.

The US is uniquely positioned to improve the safety of the global open-weight ecosystem. America is already a 

 in AI safety research; it has an enormous talent pool and hosts 

 of the world's computing capacity (safety research can require significant computational resources). America is better able to develop advanced methods for open-weight models for removing dangerous capabilities (like cyber and bio capabilities) and imposing guardrails that cannot easily be stripped out. The United States sharing new methods for safer open-weight AI with China would be akin to how it 

 with the Soviet Union and later Russia. The US would also have the capacity to quickly disseminate novel safeguards to the rest of the world through new model releases. By 

, last year, the US developed 59 significant AI models compared with 35 models for China and only one each for France and the UK.

If the US were to abandon the global open-weight ecosystem, it would only worsen the safety risks, as Chinese open-weight models proliferate in 

. And it cannot outcompete the Chinese models solely with closed-weight models, because open-weight models are 

, and the ability to modify and self-host models appeals to enterprises and researchers who want to train AI in specific tasks and keep proprietary data on systems that they control. If the US does not lead on the safe development of open-weight models, there is a risk that China will distribute 

 to every corner of the world, where they can be downloaded freely by malicious actors.

American open-weight leadership would also improve the prospects for the US and China to eventually cooperate on safeguarding and pacing new models. Open-weight models will likely require a different safety regulatory regime that accounts for the fact that users can modify the weights and developers cannot recall downloaded models. If both countries are developing frontier open-weight models, there is a symmetry that makes it easier to agree to mutual safety standards because the two governments’ commitments will be comparable. Having frontier open-weight models would also add to America’s credibility. If the US proposes a bilateral safety framework for open-weight models without having any models capable enough to be subject to the framework, the proposal risks being perceived as a pretext for kneecapping China’s AI industry.

So, what should America do to quickly build out its open-weight ecosystem?

First, the federal government should make an advance market commitment for frontier open-weight models by American developers. The government would set objective criteria, including for performance and safety, and commit to paying a certain sum to labs that develop models that meet the criteria. The government only pays 

 and does not take equity in the labs to preserve competition and limit conflicts of interest.

 to raise capital because the economics are viewed less favorably when the model weights are free. An advance market commitment could increase both private funding today and revenue in the future. But the federal investment needs to be large enough—to the tune of tens of billions of dollars.

Second, the government should build on its 

, and use procurement to give preference to secure and high-performing American open-weight models over closed models within agencies that benefit most from customized and self-hosted models.

Third, the leading closed-weight labs should consider selling 

 to American open-weight labs so the latter could pay to legally and 

 their models on the outputs of the best closed-weight models.

Skeptics may argue that the US should stay away from open-weight models for safety reasons. But ceding the global open-weight ecosystem to China would be far more dangerous.

Last week, the European Commission introduced the 

, a new legislative proposal aimed at strengthening the protection of children online.

While the initiative had long been anticipated, it remained unclear until the very end what instrument the Commission would use to complement its existing digital rulebook. By opting for a standalone regulation instead of revising existing ones, the EU executive has added another layer to an already dense regulatory landscape.

 (DSA) already contains extensive obligations to protect minors online, while the forthcoming 

 (DFA) is expected to address many of the problematic design practices targeted by the KIDS ACT.

So, the questions naturally follow: What exactly does the new proposal add? Where do these instruments overlap? And what does this growing regulatory stack mean for enforcement and the EU’s broader approach to online child safety?

Under Article 28 of the DSA, which applies since 2024, providers of online platforms accessible to minors — with the exception of micro or small enterprises — must already “put in place appropriate and proportionate measures to ensure a high level of privacy, safety, and security of minors, on their service.”

In July 2025, to give this broad obligation more concrete meaning, the European Commission published more than 60 pages of 

 detailing measures platforms may take to comply with Article 28. These range from setting minors’ accounts to private by default, to adapting recommender systems to reduce exposure to harmful content or disabling by default features that can contribute to excessive use.

Services designated by the European Commission as Very Large Online Platforms (VLOPs) or Very Large Online Search Engines (VLOSEs), such as Facebook, Instagram, Snapchat, TikTok or Google Search, carry additional obligations. 

Under Articles 34 and 35, they must assess and mitigate systemic risks to minors that may arise on their services. For VLOPs and VLOSEs, enforcement is the responsibility of the European Commission, while national Digital Services Coordinators oversee other regulated actors.

 the DSA against several major platforms used by minors in the EU. In April 2026, it preliminarily 

 Meta in breach for failing to adequately assess and mitigate the risks of under-13s accessing Instagram and Facebook. A few months earlier, it had reached a similar 

 against TikTok over addictive design features including infinite scroll, autoplay, push notifications and its recommender system. 

Four porn platforms were also preliminarily 

 in breach by the European Commission for failing to protect minors from pornographic content, while formal proceedings 

 against Snapchat concerning its compliance with Article 28 and the protection of minors.

By extending safety-by-design obligations to online games, video-gaming platforms and AI chatbots and companions, the EU KIDS ACT reaches beyond the DSA’s platform framework.

One thing the KIDS ACT adds to this already extensive regulatory landscape is turning some of the safety-by-design measures set out in the Article 28 guidelines into binding, more prescriptive rules.

Chapter III lists specific obligations for online social networking services, video-sharing platform services, online games, AI companions, general conversational chatbots, and software application stores. These obligations apply by default, unless the provider has established, through age assurance, that the user is over 18. They cover addictive design, recommender systems, safe settings, contact and interaction safeguards, economic transactions, age-appropriate access, and tools to support minors’ agency and empower them and their guardians.

Article 9 illustrates the shift: it states that social media and video-sharing platforms must not be designed or operated in ways that are intended or likely to encourage compulsive or excessive use by minors. It identifies a number of designs made to encourage compulsive or excessive use, including autoplay and endless content feeds without meaningful breaks, notifications designed to pull minors back onto a service, rewards for posting or livestreaming, and mechanisms that pressure users to return regularly or risk losing benefits.

Another major addition is the introduction of a three-tier age-based system for access to social media and video-sharing platforms (Chapter II). For social networking and video-sharing services covered by the proposal, children under 13 generally cannot create their own account. A limited exception would allow guardian-controlled access to certain age-appropriate video-sharing services designed for younger children. 

Those aged 13 and 14 can access these services only through a limited “mini account” set up and supervised by a parent or guardian, with restrictions on features, contacts and screen time.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe