On January 8, 2026, The GovLab and the Open Data Policy Lab hosted “Looking Back to the Future”—a conversation that looked at the forces that shaped 2025 and may define 2026 across data, AI, and governance.

Moderated by Stefaan Verhulst (Co-Founder, The GovLab; Research Professor, NYU Tandon School of Engineering), the webinar brought together three panelists offering different perspectives on the evolving data and AI landscape:

• Jed Sundwall, Executive Director, Radiant Earth

• Guilherme Canela De Souza Godoi, Director, Division for Digital Inclusion and Policies & Digital Transformation and Secretariat of UNESCO’s Information for All Programme

• Susan Ariel Aaronson, Research Professor of International Affairs and Director of the Digital Trade and Data Governance Hub (and Principal/Co-Principal Investigator with NSF and NIST on responsible AI)

A clear framing emerged early: While AI is experiencing a kind of “summer,” the world may also be entering a “data winter”—marked by a rapid decline in access to data for public interest purposes. While new actors might fill gaps in data access, this required updating access regimes and rethinking basic institutional foundations underlying data systems.

The 2025 Rearview Mirror: What Stood Out

1) Access to high-impact data is increasingly shaped by non-state actors

The conversation began with Sundwall highlighting how sharply declining resource and technology costs are enabling smaller nonprofits and research institutions to undertake activities that were once the exclusive domain of states. 

Sundwall pointed to the emergence of Common Space, a nonprofit established in 2025 to launch a humanitarian satellite and the Environmental Defense Fund’s MethaneSAT. Both serve as evidence that barriers to entry in space-based data collection have fallen dramatically. As he noted, these initiatives emerge in part from a recognition that “a lot of the actors who have access to high-resolution satellites and imagery and can control those things might not prioritize humanitarian efforts.”

This shift, he suggested, reflects less a deliberate move away from governments than a response to declining trust, access, and prioritization within some state-controlled data systems, with falling costs in computing, sensors, and satellite technology serving as the primary mitigating factor enabling non-state actors to step in—expanding the range of institutions capable of producing planetary-scale datasets, rather than eliminating the role of governments altogether.

2) AI can facilitate access to data—not just depend on it

Sundwall then emphasized that AI can play a role not only in consuming data, but also in making certain kinds of data easier to access and use. One example he highlighted was the use of AI-generated embeddings derived from satellite imagery—essentially compact representations of images rather than the full images themselves. As he explained, “embedding outputs trained on satellite imagery are just much, much smaller than all of that imagery that people normally have to sift through,” which means they “stand to make much more data much, much more accessible to people.”

Instead of requiring users to store, move, or analyze massive volumes of raw geospatial imagery, these embeddings allow people to work with far smaller data outputs that still retain useful information. Because they are significantly smaller, they can be easier to move, share, and use, potentially lowering technical and resource barriers to working with large-scale geospatial datasets. At the same time, Sundwall cautioned that important questions remain unresolved, including “how reliable are they?” and “what are they even telling us?” He noted that embeddings can be difficult to interpret and understand in practice.

He also pointed to a more operational role for AI: reducing the friction involved in adopting data standards. AI tools, particularly large language models that assist with writing and adapting code, can help automate what he described as “the boring chore work” of implementing standards, making it easier for more organizations to use shared standards that support interoperability and reuse.

3) Multilateral governance is undergoing “tectonic” change

Speaking from the perspective of multilateral institutions, Canela described 2025 as a year of “tectonic changes” in global governance. By this, he meant a break from long-standing assumptions that have shaped the multilateral system since World War II, particularly the expectation that states would broadly cooperate, share responsibility, and work through international institutions to address global challenges. Instead, he noted that some major state actors were increasingly stepping back from these roles or openly questioning the value of multilateral organizations themselves.

These shifts, Canela explained, made decision-making much harder. Established processes that once relied on consensus and predictable participation could no longer be taken for granted, and negotiations increasingly unfolded in a more fragmented and contested environment. As he put it, 2025 was “a very difficult moment for the decision-making process,” marked by what he described as “a clear attack against the idea of doing multilateral policy and multilateral governance.”

At the same time, Canela pointed to a more hopeful development, particularly within the UN system and regional organizations. Amid intense “AI hype” and a rush by individual agencies and institutions to launch their own initiatives, there was growing recognition that uncoordinated efforts risked duplication and confusion. Over the course of 2025, he observed signs of greater willingness among these actors to coordinate their approaches and align their work.

He highlighted the WSIS+20 review process as a concrete example of this alignment. One of the few UN General Assembly resolutions adopted by consensus, the review reaffirmed that technology governance should remain firmly grounded in human rights. Importantly, Canela emphasized that the central challenge is not how to regulate each new technology as it emerges, but how to ensure that the broader governance frameworks guiding digital transformation are rights-based, coherent, and fit for purpose. In his words, the focus should be on “the overall framework,” rather than on individual technologies in isolation.

4) Data sovereignty, national security, and cross-border flows are intensifying fault lines

Aaronson concluded the review of 2025 by highlighting two developments over the past year that illustrate diverging approaches to data governance and access.

First, she discussed China’s efforts to create formal data markets—systems intended to make buying and selling data more structured and transparent. While the idea has attracted significant policy attention, Aaronson was blunt about the results. As she put it, China’s data markets have been “shockingly unsuccessful so far” and have “not gotten a lot of companies to participate,” despite repeated attempts to make them work.

She also described China’s approach to AI governance as focusing on how AI systems are used and who uses them, rather than on the technology alone. This includes identifying specific chatbot uses considered harmful and issuing draft rules for public comment. Aaronson highlighted this as unusual in context, noting that “the Chinese government actually put the draft rules out for public comment,” which she described as “quite interesting” given the country’s broader political environment.

Turning to the United States, Aaronson emphasized a broader shift toward restricting access to data, with national security increasingly used as the justification. She pointed to a 2024 decision, made before President Trump returned to office, in which the U.S. imposed its first explicit limits on cross-border transfers of certain categories of sensitive personal data to six countries designated as “terrorist nations.” For Aaronson, this move was significant not only because of the terrorism framing, but because it marked “the first time the United States has really restricted data flows in this way.”

She warned that these measures are part of a wider trend toward reduced data access. When combined with expanding “data sovereignty” policies, she argued, they risk sharply limiting global data sharing, an outcome she described as deeply concerning, given that “science and innovation depend on data that crosses borders.”

Looking ahead: 2026 themes the panel expects to intensify

1) New data institutions beyond the state

A central expectation for 2026 is the growing role of non-state actors—such as research labs, nonprofits, and consortia—in producing datasets that increasingly function as de facto public infrastructure.

Jed Sundwall pointed to large-scale examples already emerging. One was a planetary-scale, 3D building footprint dataset produced by a research lab at the Technical University of Munich. Another was the Overture Maps Foundation, a consortium of major technology companies working to create a shared global mapping dataset and a system of unique identifiers for real-world entities.

Sundwall argued that these efforts signal a deeper shift: data critical to public decision-making is no longer produced only by governments, yet governance models have not kept pace. As he put it, “We’re going to need new kinds of institutions that can actually produce and govern data properly,” adding that these institutions are unlikely to be “purely governmental” or “purely private.”

To illustrate what he meant, Sundwall referenced existing global coordination bodies such as ICANN (which manages the internet’s domain name system) and GS1 (which governs global barcode standards). These organizations do not belong to any single government, yet they provide trusted infrastructure that underpins global commerce and connectivity. In his view, comparable institutional models may be necessary to govern shared data resources responsibly at planetary scale.

2) Rethinking “open”: from a slogan to design choices

Sundwall cautioned that “open” is often treated as a catch-all label rather than a precise concept. He described it as a sloppy or imprecise term, noting that openness is only one attribute of a data product, not a guarantee of quality, ethics, or sustainability. For clarity, he urged moving away from treating “open” as a goal in itself and instead asking more fundamental questions: Is the data useful? Is it produced and used ethically? Who pays to maintain it over time? Who captures the value it generates?

Guilherme Canela reinforced this point by calling “open” an “omnibus concept”. By this, he meant that without governance models, context, and purpose, the word risks obscuring real trade-offs rather than clarifying them. As data ecosystems expand beyond governments to include private firms, nonprofits, and hybrid actors, he argued, governance must become more explicit about how openness is designed and for whom it works.

Both speakers suggested that 2026 may mark a turning point: a recognition that producing high-quality, widely usable data is costly, and that older assumptions about unlimited openness may need revision, particularly when large, well-capitalized actors extract disproportionate value from shared resources.

3) Updating access-to-information regimes for a non-governmental data world

Looking ahead, Canela highlighted a symbolic milestone: 2026 marks 260 years since the world’s first Freedom of Information (FOI) law, adopted in Sweden in 1766. He used this anniversary to pose a forward-looking governance question: how should access-to-information frameworks evolve in a world where governments are no longer the primary holders of socially important data?

Canela urged for organizations to adapt long-standing principles—such as transparency and public access—to current realities. While roughly 140 FOI laws exist globally, he noted that only a small number meaningfully apply to private or non-state actors, with South Africa standing out as a notable example.

In 2026, he argued, the challenge will be whether access-to-information regimes can be redesigned to reflect today’s data ecosystem—one in which corporations, platforms, and research institutions play central roles—without abandoning the democratic principles that FOI laws were meant to protect.

4) Quality inputs” and the risk of a data tragedy of the commons

Canela also warned that weakening trusted sources of information could have cascading consequences across society. If science, independent journalism, and reliable data producers are undermined, he argued, the entire information ecosystem suffers—affecting not only governance but also business and innovation.

He framed this risk as a potential tragedy of the commons: When low-quality or misleading information floods the system, it becomes harder for anyone to extract real value. He linked this concern to two forward-looking signals: the approach of the final years of the 2030 Agenda, which depends heavily on data for evidence-based policy, and the inclusion of “information integrity” in the political outcome of COP30 for the first time. For Canela, 2026 will test whether these commitments translate into concrete action.

5) Responsible AI: transparency without accountability

Susan Ariel Aaronson finally emphasized that debates around responsible AI are likely to sharpen in 2026, particularly around enforcement and accountability.

She argued that many corporate commitments to responsible AI are inconsistent in practice. Drawing on her own research, Aaronson described comparisons between what companies claim in technical documentation and public statements and how their chatbots actually behave over time, especially on issues related to user welfare (such as responses to self-harm) and user rights (such as data protection and freedom of expression). While some systems improved, others showed gaps or reversals, suggesting that voluntary commitments  alone may be insufficient.

Aaronson also highlighted the challenges facing the Independent Evaluators Forum, an emerging effort to assess AI systems. The forum’s central problem, she noted, is limited access to training data and evaluation inputs, which makes credible, independent assessment difficult. Without some form of mandated disclosure or cooperation, she argued, transparency cannot easily translate into accountability.

More broadly, Aaronson warned that a public “wake-up on data” has yet to occur. Even where laws exist, such as data protection rules, enforcement gaps persist, including around practices like web scraping. For her, 2026 will hinge on whether governance mechanisms move beyond aspirational transparency toward enforceable accountability.

Inclusion beyond access: language, participation, and the digital divide

The conversation also returned to the unresolved challenge of inclusion. Canela underscored that the divide is not only about the 2.6 billion people still offline, but also about meaningful participation for those who are connected. He highlighted the scale of linguistic exclusion: with roughly 8,000 languages spoken globally, only a fraction can meaningfully interact with digital ecosystems in their own language—suggesting a much deeper participation gap than connectivity statistics alone capture.

Closing reflections: three words for 2026

In the final minutes, Verhulst asked each panelist to name one concept likely to define the year:

• “New data institutions” (Sundwall)

• “Cooperate or isolationism” (Canela)

• “Manipulation”—specifically “neural manipulation” (Aaronson), including concerns about AI agents influencing users in ways people may not recognize

Key Takeaways

Across perspectives, the webinar surfaced a shared set of tensions that will likely intensify in 2026:

• Data governance is increasingly shaped by actors beyond the state, raising the need for new institutional models.

• “Open” needs to be treated as a choice within governance design, not a substitute for strategy, sustainability, or ethics.

• Quality, accountability, and information integrity are becoming central constraints—especially in an environment where traditional accountability institutions face pressure.

• Inclusion requires more than access: language, context, and meaningful participation must be part of the governance agenda.

As Verhulst noted in closing, the discussion was intended not as a final word, but as the basis for a research agenda—and an action agenda—for the year ahead.

Cover image from Pixel Particle from Getty Images Pro

Back to the Blog