A yard sign expressing opposition to the construction of a data center is seen outside of a home in Mifflinville, Pa. on July 3, 2026. (Photo by Paul Weaver/Sipa USA via AP Images)

Data Center Discontent Drives State Legislation Surge

L to R: Open AI CEO Sam Altman, US President Donald Trump, CEO of DeepMind Demis Hassabis, South Korea's President Lee Jae Myung and German Chancellor Friedrich Merz attend a working session during the G7 Summit in Evian, France, on June 17, 2026. (Jeanne Accorsini/Sipa via AP Images)


AI Systems Are Getting More Powerful. The Ability to Verify Must Keep Pace.

The Capitolio Nacional in Bogotá, where Colombia's Congress is considering Bill 025 to regulate artificial intelligence.

Colombia is Preparing a Poor Copy of the EU's AI Act

French President Emmanuel Macron. (Manon Cruz/Pool Photo via AP)

France’s Constitutional Council Ruling Tests the Limits of Social Media Bans

United States Treasury Secretary Scott Bessent speaks to members of the media outside the White House in Washington, Thursday, Aug. 20, 2026. (AP Photo/Jacquelyn Martin)

What the Bond Market Disruption Means for Rebuilding American Tech Policy

Indian Prime Minister Narendra Modi addresses the nation from the rampart of the 17th century Mughal-era Red Fort monument during the country's Independence Day celebrations in New Delhi, India, Saturday, Aug. 15, 2026. (AP Photo)

Technocracy Won't Debug India's Education Crisis

Saline, Michigan, December 1, 2025: A sign on a rural Michigan road opposes a planned $7 billion data center on southeast Michigan farm land. Opponents say the data center could raise residential electricity rates and endanger the water supply. (Photo by: Jim West/UCG/Universal Images Group via Getty Images)

What to Do With a Data Center Pause — and Unpacking the ‘Censorship-Industrial’ Complex Narrative

House Judiciary Committee Chairman Jim Jordan, R-Ohio, conducts the Select Subcommittee on the Weaponization of the Federal Government in the Rayburn Building on Thursday, July 20, 2023. 

How the 'Censorship-Industrial Complex' Narrative Went From Fringe to Trump Policy

Abstract image of a whistleblower amid a see of faces.

Congress Can Get AI Whistleblower Protection Right On Its First Try

Nine small images with schematic representations of differently shaped neural networks, a human hand making a different gesture is placed behind each network.

AI Audits Need a Power Test, Not Just a Fairness Score

The Mathare River is one of the rivers that flows into Nairobi River. Photo by Queen Asali/Wiki For Sustainable Futures 2026/CC by 4.0

The Mathare River is one of the rivers that flows into Nairobi River. Photo by 

Taxing Tech in Age of AI Could Fund Human Rights

Composite. Left, Supreme Court Justice Samuel Alito attends a meeting in Rome, Sept. 20, 2025. (AP Photo/Andrew Medichini) Right, Meta CEO Mark Zuckerberg, CEO of Meta, is seen in the US Capitol on Thursday, March 26, 2026. (Tom Williams/CQ Roll Call via AP Images)

How the Supreme Court and Big Tech Both Weaponize 'Race-Neutrality'

Children’s Digital Rights Need More Than a False Promise of Safety

In an aerial view, the Meta El Paso Data Center is seen on August 13, 2026 in El Paso, Texas. Meta continues construction on its more than $10 billion data center, occupying approximately 1,000 acres, and has expanded the project to 1 gigawatt of computing capacity, allowing the facility to support a growing range of AI workloads and the training and deployment of AI models and technologies. (Photo by Brandon Bell/Getty Images)

Zuckerberg’s AI Manifesto Lost it at Compute

Flock Safety camera on Haverhill Drive near Phillips Avenue in Toledo, Ohio. (Photo by Stephen Zenner / SOPA Images/Sipa USA)(Sipa via AP Images)

Unless Flock is Reined In, Mass Video Surveillance Will Grow Even More Pervasive

Protesters rally against data centers in Hillsboro, Ore., on May 16, 2026.  (Photo by Alex Milan Tracy/Sipa via AP Images)

How to Use Your Community’s Data Center Moratorium

Chinese President Xi Jinping, center, waves as he arrives at the opening ceremony for the World AI Conference in Shanghai, Friday, July 17, 2026. (AP Photo/Ng Han Guan, Pool)

Will China Crack Down on Open-Weight Models?

Brazilian presidential candidate Ronaldo Caiado, former governor of Goias state, presents his public security policy in Sao Paulo, Brazil August 10, 2026. REUTERS/Jean Carniel

How a Brazilian Governor Turned Rare Earths Into a Campaign—and Washington Into a Partner

CEO of Meta Platforms Mark Zuckerberg attends the Allen & Company Sun Valley Conference at the Sun Valley Lodge on July 9, 2026 in Sun Valley, Idaho. (Photo by Kevin Dietsch/Getty Images)

Four States Take Meta to Trial in California, the Latest Front in Sprawling Multidistrict Fight

Vietnam Tightens Online Speech Controls as Cyber Law Takes Effect

As European member states such as France, Austria, Greece, and Spain 

 establishing social media bans for children and minors of varying ages, the 

 of the French Constitutional Council, the country’s highest constitutional authority, is a clear signal and reminder of the primacy of ensuring policy centers on, and at the very least does not impede, the fundamental rights of children and all users.

On Aug. 14, France’s Constitutional Council, the country’s highest constitutional authority, struck down the central provision of a law that would have prohibited children under 15 from accessing social media. The law, 

 on July 21, was intended to address risks the Parliament determined were associated with social media use, including anxiety, depression, sleep disorders and online harassment. It was due to take effect at the start of the 2026-27 school year.

Now, the Constitutional Council has declared Article 1 of the bill 

. The Council did not reject the objective of protecting children. Instead, it found that the blanket prohibition was disproportionate because it did not adequately account for differences among services or among individual children. 

It also found that the legislation failed to provide sufficient legal safeguards for age assurance, which could require adults as well as children to prove their age before accessing covered services. President Emmanuel Macron has tasked Prime Minister Sebastien Lecornu to prepare a revised proposal, with the government aiming for a new version by spring 2027. 

 (Law aiming to protect minors from risks associated with social media usage), was expected to be a sweeping victory for President Macron’s final year in office. 

We discuss three key ideas from the French Constitutional Council’s decision that should be considered as the institutions develop their EU-wide approach on child online safety.

Broad scope of services covered violates freedom of expression

One reason the French Constitutional Council revoked Art. 1 of the bill is its failure to distinguish between the services concerned based on their nature, functionalities, risks, and needed safeguards.

The Council found that the ban, defined by cross-reference to the EU's Digital Services Act and Digital Markets Act via Article 6 of the French law of 21 June 2004, "covers access to any online platform allowing end users to connect and communicate with one another, share content and discover other users and other content," and that access to such services "is not subject to any condition relating to the functionalities or content offered, the dangers to which they expose users, and the inadequacy of the safeguards with which they are provided.”

It further noted that the prohibition carried some exceptions. It did not apply to online encyclopedias, educational or scientific directories, or platforms for the development and sharing of free software or open-source digital projects with an educational purpose. However, the Council found that these exemptions "remained limited.”

Although the Council agreed in principle that protecting children's best interests may justify restricting minors’ access to social media, it found the ban neither necessary nor proportionate to the objective pursued, and unduly infringing on children's and minors' freedom of expression and communication. As enshrined in 

 of the EU Charter of Fundamental Rights, freedom of expression is essential for children and minors, even if it is often overlooked.

, and many civil society organizations, the Council noted that broad social media bans disproportionately deny young people’s rights to access to information and democratic participation. 

Indeed, children are active citizens who need to participate in their nations’ democratic processes. Excluding minors below the voting age from accessing a wide variety of social media services and, hence, from participating in online debates around elections is a missed opportunity to prepare and equip future voting citizens with the knowledge of how they can meaningfully engage in debates on democratic values in the modern age, which largely happens on these services.

By treating the entirety of social networking service(s) as defined by French law (which uses the definition of 

) as a monolithic threat, the bill overblocked benign digital infrastructure, ignoring the reality that digital platforms are also essential tools for education, civic engagement, and community building. 

 accompanying the Constitutional Council’s decision, Art. 1 indiscriminately restricts access to educational, cultural, and civic content, as well as the exercise of freedom of expression by young people aged 13 to 15 and the interpersonal communication tools that are essential to modern social life.

The French Constitutional Council ruled that the lack of safeguards for the deployment of age verification falls short of constitutional privacy guarantees for all users, adults and children alike, in France. 

[by] prohibiting any minor under the age of fifteen from accessing certain online services, the contested provisions themselves imply that every person, even an adult, must prove his or her age before accessing them. In the absence of a determination of the conditions and limits within which this must thus be justified, the legislature has not provided the legal safeguards of such a nature as to ensure compliance with these constitutional requirements.

Considering the scope and scale of privacy risks inherent to widespread age assurance, it is surprising that the legislature has not seen the need to add clear safeguards for age assurance. 

 to the Council on Aug. 5, 2026, it appears that this was not an omission, but rather intentional. The government argued that the Digital Services Act (DSA), in particular the Art 28 

, conclusively regulates age assurance, and that the French legislature hence does not have the competence to set (additional) safeguards and parameters for the deployment of age assurance: ‘

The definition of the prohibition – access to social media services for minors under 15 – and the definition of the arrangements for compliance therewith, its monitoring and, where applicable, its sanction

 accordingly fall within two different bodies of rules – the first, domestic law, the second, European Union law.’

The Council did not engage with this argument and, as cited above, plainly stated that the missing safeguards in the national legislation fell short of constitutional requirements. The reason for this might be that the safeguards and parameters highlighted by the French government are not included in the DSA itself, but merely in the 

, which are non-binding. It is the Guidelines that recommend which age assurance method should be deployed for which safeguarding purpose, and how the appropriateness and proportionality of age assurance should be assessed in terms of its accuracy, reliability, robustness, non-intrusiveness and non-discrimination.

As these parameters are only included in the Guidelines, and not in the DSA as the underlying legislative act, it can be argued that the EU has not conclusively ‘regulated’ age assurance. In the context of a potential social media ban, the French government could have incorporated the decision whether to deploy age assurance as well as the safeguards for such deployment contained in the Guidelines into any national legislative act. Such an inclusion would not constitute an infringement of the EU’s competences, but rather a manifestation of non-binding EU guidance at national level.

When finding the law to be constitutionally deficient, the Council focused only on the lack of legal safeguards in the national law. However, this does not mean that including safeguards would automatically cure the unconstitutionality of widespread age assurance. Depending on the scope and method of deployment, the widespread deployment of age assurance might still carry significant risks to privacy, non-discrimination, freedom of expression and access to information. 

As the Council did not provide any guidance on what legal safeguards are required to ensure the ban meets privacy constitutional guarantees, the legislature has to reconsider its approach and develop such legal safeguards for the next iteration of the law.

Lack of parental oversight compounds the freedom of expression violation

The Council held that the exclusion of parental oversight as a mechanism to override or limit the ban undermines the right to freedom of expression. In the eyes of the Council, this exclusion of parental participation in the child’s social media consumption leads to the complete disregard for assessing the individual risk to a minor, 

‘taking into account in particular his or her age, degree of maturity and family situation, as well as the nature of the service concerned.’

, stating that a total ban institutes a general presumption of ‘educational incapacity’ against all families in France. 

This was further supported by ONEST in their 

[i]t is for the parents, not the State, to assess, child by child and service by service, what the age and maturity of the one permit with regard to the other

The Council therefore appears to give a clear directive to the legislature to include an individual assessment mechanism, possibly through parental decision-making power, in the next iteration of the law, rendering an ‘absolute’ ban unconstitutional. The decision is ambiguous in this regard, criticizing on the one hand the lack of parental involvement in governing children’s access to social media, while on the other hand stating more broadly that the lack of any particular assessment of the risk to the minor poses a disproportionate infringement of the right to freedom of expression.

The Council underlined that legal guardians should be able to "decide to lift the ban, to limit its scope or to authorize access to certain services" in the interest of the child. If read as a call for a parental decision-making mechanism, this might, in theory, allow social media access for those under 15 in cases where they appear sufficiently mature and resilient to respond to potential risk encounters.

However, implementing a parental decision-making mechanism will depend on establishing some form of verifiable parental consent (VPC), including a mechanism to determine whether a user is a child and to accurately link the identity to the child's legal guardian, depending on the degree of certainty required. This poses a myriad of legal and factual implementation issues.

experience with the protection of children’s personal data

 under the General Data Protection Regulation, service providers genuinely struggle to provide a VPC mechanism that verifies that the person giving consent on behalf of the child is indeed the child’s parent or legal guardian. 

serious concerns with regards to reliability and equity.

 Approved techniques such as government ID checks, facial recognition matching, knowledge-based authentication, and credit card verification are each susceptible to circumvention by children, carry significant privacy risks for parents, and impose enough friction that many families do not complete them.

There is no solution for cases in which children’s parents have separated and both have different opinions about consenting to their child’s use of digital services. Common VPC methods exclude millions of caregivers who are unbanked, undocumented, or from non-traditional family arrangements. The burden falls disproportionately on the most vulnerable children, i.e., those in foster care, or with absent or disengaged parents, who are arguably those the policy is most intended to protect.

This shows that parental consent, instead of a blanket ban, appears to be less intrusive in theory, but poses serious questions in terms of effectiveness, reliability and equity.

Council ruling sets the tone as Brussels weighs an EU-wide social media ban

legislative proposals by EU member States

 on the protection of children and minors online, the Constitutional Council’s decision has sent a shockwave expected to have some effect in Brussels ahead of the highly anticipated 

 on child online safety due by December 2026. The decision exposes a critical regulatory flaw that has remained overlooked until now: penalizing children and minors with bans, while politically expedient, is shortsighted regarding the fundamental rights implications and will likely face significant legal and practical limitations.

Age assurance deployment is a core part of the EU’s approach to child online safety, even outside a general ban. Alongside ensuring all service providers comply with the Article 28 guidelines, providers should conduct a Fundamental Rights Impact Assessment (FRIA), including a Data Protection Impact Assessment (DPIA), before deploying any age assurance measure. It is essential to ensure that both the decision whether to deploy age assurance and the choice of the means of deployment remain strongly grounded in fundamental rights.

National bans mainly targeting users rather than platforms themselves miss key opportunities to address the root causes of many of the concerns related to minors’ activities on social media. 

Some of these may be addressing and systematically identifying risks that may be posed by design features present on services and identifying pathways to mitigate such risks. Encouraging platforms to change their addictive design features under the DSA and Article 28 guidelines may achieve the desired objective of automatically protecting minors and children as well as protecting all users online without requiring invasive, state-mandated digital exclusion and overreach.

It is now clear that platforms are increasingly facing court cases seeking to hold them responsible for design features of their products worldwide, with the drumbeat of court cases in the US, including some initial rulings against platforms. 

Notwithstanding whether they will go unchallenged or their effectiveness, efforts like the Art 28 Guidelines issued under the Digital Services Act may actually shift company incentives to identify and mitigate risks posed by these design features beyond just hitting companies where it will have the most impact on their economic incentive. However, shifting corporate incentives is just one side of the coin. Ensuring that existing and upcoming regulatory mechanisms strike the right balance and withstand judicial scrutiny is the other side.

What policymakers could take from the ruling

As the 2027 French presidential race begins to ramp up, the Constitutional Council’s decision sets a unique precedent by creating a high-stakes campaign subject that candidates across the political divide will likely exploit, potentially framing the rejection as either an overreach of judicial activism or as a sign of executive incompetence in drafting enforceable laws. 

Consequently, President Macron’s push for a revision before the Spring of 2027 positions the Bill as a formidable test of boundaries between state-mandated digital protection and constitutional civil liberties. Gabriel Attal, a French political figure, has already demanded a 

, using the Constitutional Council’s block to frame mainstream politics as disconnected from everyday parental anxieties and restrained by elite judicial oversight.

Looking at Brussels, the Constitutional Council’s decision should force a much-need reflection moment on the necessity for the European Commission to grapple with the extent of alignment between its anticipated EU-wide framework and the protection of fundamental human rights of minors and children as well as of all users. 

In the immediate term, the Constitutional Council’s signal is crystal clear: Protecting minors and children online cannot be achieved by infringing on their rights to freedom of expression, access to information, privacy, and data protection. With that in mind, EU policymakers must pivot from restrictive access control to insist on platform safety-by-design and by default.

. He established the US Center for AI Standards and Innovation at the National Institute of Standards and Technology, and from 2017 to 2020, he served as the first assistant director for Quantum Information Science at the White House Office of Science and Technology Policy.

L to R: Open AI CEO Sam Altman, US President Donald Trump, CEO of DeepMind Demis Hassabis, South Korea's President Lee Jae Myung and German Chancellor Friedrich Merz attend a working session during the G7 Summit in Evian, France, on June 17, 2026. (Jeanne Accorsini/Sipa via AP Images)

As the White House debated its approach to AI policy this summer, headlines demonstrated both the promise and the peril of the technology. A series of disclosures that systems developed by 

 each hacked third-party systems during testing raised alarm, while news that an OpenAI system produced 

 accompanied by proofs that could be independently checked impressed (and frightened) mathematicians. This series of events makes clear that Washington’s AI policy requires a key pillar: powerful systems require consistent verification against safety benchmarks that are public and interpretable.

AI capabilities are growing rapidly, but the capacity to check these systems is not. The resulting gap — verification asymmetry — is widening.

Closing it requires public, standardized testing tools with integrated formal reasoning checks, built where the work is happening, as it is happening. The Center for AI Standards and Innovation (CAISI) at the National Institute of Standards and Technology (NIST) should publish a public tier of frontier model evaluations alongside the classified process, built by engineers embedded with the teams shipping these systems.

Semiconductor manufacturing faced a structurally similar problem and answered it with in-line metrology. You can test a wafer at the end of the production line and learn only that it failed, or you can measure at each step and learn which step failed while you can still act on it. Frontier AI evaluation is largely an end-of-line test today, applied to a process that ships every six months.

When I helped establish the federal AI safety effort at NIST in 2023, our 

 came from an executive order written for a generation of models far less capable than those deployed today. We built it integrated with a consortium, and that part worked: more than 200 organizations at launch formed an active community of practice around child safety, biodefense, and provenance for synthetic material — norms that moved through the participating companies and, through them, across the industry.

What we got wrong was placement. We needed evaluation teams working inside and alongside the frontier labs, and the small, excellent team we assembled had neither the funding nor the access to do it. That gap has not closed, and the systems have grown considerably more capable since. In perhaps the most significant of those headline-generating events this summer, a system using OpenAI’s models while running a cyber-capability evaluation found a 

 and used it to get inside Hugging Face's production infrastructure. Hugging Face detected and contained the intrusion five days before OpenAI connected it to its own testing.

Fortunately, progress is being made. An unreleased OpenAI model 

 10 open problems in mathematics and theoretical computer science that included machine-checkable proofs. Anyone can download those proofs and verify them in minutes—verification in mathematics works. But in critical domains, such as engineering, biotechnology, and defense, we are lagging.

's classified cyber-capability benchmark for defining a "covered frontier model" is a good first step. The alternative on offer is industry self-regulation. DeepMind CEO Demis Hassabis has 

a standards body modeled on the Financial Industry Regulatory Authority (FINRA) and funded by the frontier labs, testing models the companies volunteer up to thirty days before release. That places the measurement inside the institutions being measured. But we need public, verification-based definitions across the landscape of applications, and we need them now in domains where mistakes have physical, economic, or national-security consequences.

Domain experts are already moving publicly in this direction: Harmonic and the American Institute of Mathematics 

 the development of an openly published, mathematician-designed benchmark, chosen because those answers are formally verifiable.

CAISI should scope a public tier this year, starting with the domains where the NIST consortium already has working communities—biodefense, child safety, and provenance for synthetic material. Agencies should pilot verified purchasing in a handful of programs. And Congress should fund the embedded teams that 2023 left out.

Federal purchasing is the fastest lever available. Federal agencies are among the largest buyers of AI systems for critical infrastructure, logistics, health care, and defense, including the assistants now embedded in ordinary office software. Agencies can require that a delivered system carry evidence produced against the public tier: which constraints were checked, by what method, and within what uncertainty bounds. Pilot solicitations in a handful of programs would establish the practice. Federal Information Processing Standard (FIPS) 140 

 built the market for trustworthy cryptographic modules this way, and NIST has run it since 1995.

The objection to publishing what we test for is that it shows an adversary what to evade. Cryptography settled this decades ago: the algorithms protecting federal systems are public, and NIST ran post-quantum standardization in the open, which is why the standards are trusted. Engineers are willing to do this work; published methods and reproducible measurements are what keep it independent of the institutions they come from. A public tier at CAISI would let the rest of us check the work.

Measurement that lags the systems it measures describes a world that has already moved on. The United States, through CAISI, can publish evaluations that answer three questions for developers, agencies, and the public: what a system does reliably, where it fails, and how much remains unmeasured. Congress can fund that work, agencies can buy against it, and the companies building these systems can staff it with engineers who know the terrain. As a nation, the US will keep building more capable systems. Government must support this effort by providing the means to check at the speed of progress, enabling trust to come alongside innovation instead of years behind it.

 after a record that drew criticism over security, health, public finances and, above all, its failure to turn science, technology and innovation into a genuine state priority. Before leaving office, however, its congressional caucus moved to address an issue it had not pursued with sufficient urgency during four years in government 

by introducing a bill to regulate artificial intelligence in Colombia

Bill 025 of 2026, currently before the Colombian House of Representatives

, seeks to regulate AI and ensure its ethical and responsible development. The bill applies to developers, providers and deployers of AI systems and establishes specific duties for systems that may significantly affect fundamental rights or other protected interests. These include risk and impact assessments, transparency, human oversight and monitoring, alongside public oversight responsibilities and sanctions for non-compliance. 

There is only one problem. The country does not need another ambitious law that looks good on paper but ignores the technological, institutional and economic conditions in which it would have to operate.

The bill’s purpose is legitimate. Colombia needs rules for a technology that is already transforming work, education, public administration, security and the exercise of fundamental rights. At first sight, the bill follows a familiar international approach. It adopts risk-based regulation, establishes obligations for certain AI systems and proposes mechanisms for impact assessment, oversight and governance. 

Much of this resembles the architecture of the European Union’s Artificial Intelligence Act. But there is a fundamental difference between learning from the European model and attempting to reproduce it.

The EU AI Act did not emerge in a vacuum. It operates within an integrated market of hundreds of millions of people, supported by national authorities, substantial administrative capacity, technical expertise and significant financial resources. It is also part of a broader legal and institutional system capable of distributing responsibilities among developers, providers, deployers and regulators. Colombia can adopt European principles. What it cannot simply import are the institutional capacities that make those principles enforceable.

. A framework designed for one institutional and economic environment can lose effectiveness when transferred to another without sufficient adaptation. In Colombia, the risk is particularly serious. A sophisticated regulatory framework could create obligations that local companies and public institutions are unable to meet while doing little to constrain the foreign companies that actually develop the most powerful systems. That would not be effective regulation. It would be regulatory mimicry.

Among its key proposals, the bill would require impact assessments, establish oversight mechanisms and create institutional structures for AI governance. These measures may be reasonable in principle. What is far less clear is who would implement them, how responsibilities would be distributed and with what resources. This is not simply a question of public finances. It is a question of state capacity.

The comparison with the EU AI Act is particularly relevant in the allocation of responsibilities. Under the European framework, providers of high-risk systems are responsible for documentation, risk management and demonstrating compliance, while deployers are primarily responsible for how those systems are used, including human oversight and monitoring. This distinction matters in Colombia, where a local company may deploy a high-risk system developed abroad without access to its underlying model, training data or internal documentation. Reproducing these obligations without considering who actually controls the technology could place the greatest burden on those least able to meet it.

Regional human rights law has long recognized a difference between formally recognizing a right and creating the conditions necessary to guarantee it. See, for instance, 

 before the Inter-American Court of Human Rights. A legal obligation that cannot be monitored, investigated or enforced risks becoming little more than a declaration. The same applies to AI regulation. It is not enough to require impact assessments or algorithmic audits if the authorities responsible for supervising them lack the necessary technical expertise, infrastructure, information and budget. 

Regulating technology requires much more than passing legislation. It requires officials who understand technological systems, inspection mechanisms, sanctions and cooperation with foreign regulators. Otherwise, Colombia risks creating a law that exists mainly on paper.

There is another problem that deserves greater constitutional attention. Classifying an AI system as high risk is not merely a technical decision. That classification can determine which legal obligations apply, what uses are restricted and, in some cases, whether a system can be deployed at all. That makes it important to ask who defines the criteria and under what safeguards.

From the perspective of algorithmic due process, decisions that may affect fundamental rights should be transparent, open to challenge and subject to effective review. 

 establishes the categories and general criteria for risk classification, but also allows the national AI authority to establish and update the list of high-risk uses through a reasoned administrative act and after public consultation. Article 7 designates the Ministry of Science, Technology and Innovation as that authority and gives it power to issue binding technical recommendations on the risk level of AI systems.

Although the bill reserves final intervention measures to the authorities with relevant legal powers, this raises questions about how much discretion an executive authority should have in determining which systems trigger enhanced legal obligations. Technical expertise is indispensable, but decisions with significant consequences for fundamental rights should remain subject to clear legislative criteria and effective review.

There is a deeper problem. Colombia may require a company to explain, audit or assess an AI system that it uses. But what happens when that company does not actually control the technology?

A small Colombian company may rely on a model developed by a foreign company. It may have no access to the model's training data, architecture, evaluation procedures or internal safeguards. It can deploy the technology, but it cannot meaningfully govern it. This reveals one of the great asymmetries of today's technological order. Countries in the Global South increasingly consume technological infrastructures developed by companies concentrated in a handful of countries. They may regulate how those systems are used without having meaningful control over the systems themselves.

Colombia could therefore end up imposing the heaviest regulatory obligations on those who are closest to the consequences of AI, rather than on those who possess the greatest capacity to shape the technology. This is where the debate over technological dependency and the coloniality of data becomes relevant. The issue is not simply who owns the data. It is also who controls the infrastructure, the models, the computing power and the knowledge required to transform data into technological and economic power. A genuinely sovereign AI policy should confront that distribution of power rather than simply reproduce regulatory categories developed elsewhere.

None of this means that Colombia should refrain from regulating AI. It means that regulation should be proportional to the risks and compatible with the country's actual institutional capacity. The strongest safeguards should initially focus on areas where AI can cause particularly serious harm to human rights, including health, justice, security, surveillance and public services. Obligations should also vary according to the level of risk and, crucially, according to the degree of control each actor actually exercises over the technology.

It makes little sense to require a small Colombian company to guarantee the explainability of a closed model whose operation it cannot inspect. The bill places obligations on actors involved in the development and implementation of AI, but does not establish the same clear extraterritorial reach as the EU AI Act over foreign providers. This could leave local deployers carrying obligations over technologies whose underlying models, data and documentation remain outside their control. The objective of AI regulation should not be to look European. It should be to work in Colombia.

The Historic Pact's proposal starts from a legitimate concern, but it appears to confuse regulatory sophistication with regulatory capacity. Colombia does not need to copy the EU AI Act. It should adopt its risk-based approach and clearer allocation of responsibilities between providers and deployers, while adapting compliance requirements to Colombia's institutional capacity and technological ecosystem. The more fundamental question is where the technological power that this law seeks to regulate actually resides.

The future of AI governance will not be determined only by what machines are allowed to do. It will also depend on who has the power to build them, who controls the infrastructure on which they operate and who has the real capacity to impose limits on them. That is the regulatory question Colombia should be asking before it copies Europe's answers.

A promotional graphic for the 2027 Tech Policy Press fellowship program that includes the deadline and the techpolicy.press/fellowship URL.

2027 Tech Policy Press fellowship applications are due on September 24, 2026.

Tech Policy Press is now accepting applications for its 2027 fellowship cohort.

 in 2023, the fellowship program has supported 25 fellows from 11 countries across three cohorts, who have written on pressing topics such as the governance of artificial intelligence, tech regulation, digital rights, and the entanglement of state and corporate power. You can learn more about past fellowship cohorts and read their work 

For 2027, we intend to support a new cohort of fellows to pursue independent reporting and analysis that explore urgent questions at the intersection of technology and democracy. The year-long, part-time fellowship is designed to support mid-career journalists, researchers, and public policy professionals. We welcome a wide range of contributions, from original reporting to policy analysis.

Applicants should come prepared with a clear area of focus and a sense of how their work will contribute to informing both the public and decision-makers. We are particularly interested in projects that center communities and regions often excluded from global tech policy conversations.

Each fellow will receive a $10,000 stipend to support their work. In addition, fellows will join an international cohort for regular feedback and exchange of ideas, receive ongoing editorial guidance, and participate in monthly sessions designed to support and strengthen their work.

Applicants should have a strong background or expertise in a specific tech policy area or issue, a particular geography, or, more generally, technology and its impact on democracy or society. All fellows will work remotely as independent contractors.

Contribute (on at least a monthly basis) analysis or reporting on issues related to technology and democracy. (While topics can change after the fellowship commences, each fellow should clearly articulate a topic and why they are a good fit to cover it)

Attend a mandatory hour-long monthly cohort meeting

Attend a monthly check-in meeting with Tech Policy Press’s editorial team to evaluate progress and discuss next steps

Collaborate with other fellows and the Tech Policy Press masthead as appropriate

Meet all agreed-upon deadlines for outputs

Disclose any potential conflicts of interest

If you would like to apply for the fellowship, complete 

 by 11:59 p.m. EDT on September 24th, 2026

Cover letter (no more than two pages) that conveys what you hope to accomplish

We will host two information sessions for potential applicants via ZOOM. Registration is required. Choose from one of two dates:

The fellowship is open to journalists, researchers, lawyers, and public policy professionals who are interested in exploring the intersection of technology and democracy. We encourage applicants from the Global Majority and underrepresented communities in media. Candidates should have a demonstrated interest or background in tech policy, media, or related fields, and be committed to producing high-quality, public-interest work during the fellowship.

What are the benefits of being a Tech Policy Press fellow?

The opportunity and support to work on a pressing, underreported issue for a year

Develop a significant body of work by the end of the fellowship

Access to an international cohort of past and current fellows

Engage in learning opportunities provided by Tech Policy Press

Develop and maintain a network of sources in your focus area

Can you provide a timeline for the application review process and when selected fellows will be notified of their acceptance?

We aim to notify applicants of their status by early December 2026.

When does the fellowship start and how long does it last?

The fellowship will commence in January 2027 and last for 12 months.

Are you accepting international applicants?

Can I share additional information that is not part of the application questions?

We will only assess applications based on the information required in the application form.

A published link should direct to work that is publicly accessible online and clearly attributed to you. This can include reported pieces, policy or issue analyses, opinion articles, essays, podcasts, newsletters, or data projects. Please note that social media posts and unpublished material, such as class assignments or dissertations, will 

Will there be an interview round for the fellowship?

A select pool of candidates may be invited for an online interview.

How does Tech Policy Press plan to promote and showcase the work of fellows to a wider audience?

We publish fellows’ work on the Tech Policy Press website and feature it prominently in our newsletter to reach a broad and engaged audience. Additionally, we highlight fellowship projects through our podcast and at special events throughout the year.

Is the 2027 cohort open to those whose main contributions are technical?

Yes. We welcome applications from a variety of disciplines, as long as you can show how your work aligns with Tech Policy Press’s editorial purview.

Can we upload pre-print papers or research papers that are not formally published yet?

What kinds of projects have past fellows worked on, particularly for candidates from policy, governance, or development backgrounds?

You can learn more about our fellows and their work via the 

Are there any preferred geographic or thematic focuses for applicants from the Global South?

What are some of the criteria to reject an application?

Applications may be rejected if they do not meet eligibility criteria as listed on our 2027 Fellowship announcement.

Yes, but we recommend submitting at least one solo publication.

You can share translated work, but we also require at least one submission to be written in English.

Should the monthly submissions be exclusive to Tech Policy Press?

Is this fellowship open to candidates outside the USA who are not authorized to work in the USA?

Yes. This fellowship is open to candidates from around the world.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe