Over the last three years, a global wave of laws has settled on one dominant tool to deliver that protection: mandatory age verification for online platforms. The UK’s Online Safety Act opened this door in 2023. Australia has since banned under-16s from social media outright. France passed the first EU-based law, just to have it sent back by the Constitutional Court for fundamental right issues, although it did exclude online educational resources. Laws in Spain, Austria, Greece and Italy, just to list a few, are in the making. And of course, after an expert panel report,, everyone is expecting a move from the European Commission. 

Nobody really disputes that children need protection online.The instinct behind restricting access to some online spaces is understandable and sometimes correct. But such a tool should be used on the highest-risk corners of the internet, not to be applied too broadly.

Consider what age verification actually means. Even with the most sophisticated and privacy preserving systems a platform that doesn’t collect user data will know more about you than before. Even if it doesn’t want to. And it would mean that you, regardless of age, can’t just directly access that article about Ithaca. It is an additional barrier to access to knowledge. This is why the purpose and risk of an online space need to matter when lawmakers design these rules. Something the French Constitutional Court seems to agree with. 

Wikimedia takes child protection seriously: alongside regular DSA risk-assessment and mitigation obligations, the Wikimedia Foundation has published a Child Rights Impact Assessment, scans the projects for known child sexual abuse material, and works with a global volunteer community that decides on and enforces policies. Wikimedia Europe invests in child safety training for the volunteer editors. Protecting children is not a box we tick reluctantly; it is part of our human rights commitments.

Structurally, Wikipedia and its sister projects lack features that make other platforms genuinely high-risk for children: there is no algorithm pushing people to keep scrolling, no advertising or data-monetisation model, no private messaging between users. Content is transparently, publicly moderated by a volunteer community. That doesn’t make them risk-free, but it is a fundamentally different model from commercial social media feeds. It is lower risk. There are other such spaces online, beyond Wikimedia. They deserve to be easily accessible.

Recommendations for DSA revision

A clear definition of disinformation

The DSA – and EU legislation more in general – does not contain a legal definition of disinformation or of election disinformation nor does it include a specific provision on the removal of disinformation. While the recent European Media Freedom Act (EMFA) refers to disinformation as content that is legal, but harmful, several Member States introduced laws criminalising disinformation. This means that under the DSA, disinformation could be qualified as “illegal content” and ordered to be removed. This could bring about some friction with the fundamental right to freedom of expression and the European Court of Human rights jurisprudence on Article 10 ECHR. The Commission should address this tension in the next revision due by November 2027.

The DSA should not treat Wikipedia as commercial online platforms

The definition of online platform in the DSA is rather broad and includes Wikipedia. Narrowing its scope by excluding platforms that do not operate “on the basis of personal-data collection and -monetisation, algorithmic systems, advertising, and personalisation” would mean to “better capture meaningful differences in platform risk profiles”. 

The Commission should therefore rethink the criteria according to which online platforms are designated as VLOPs and introduce a “more risk-sensitive designation model”, and to consider “whether an encyclopaedia exception should be included”. Furthermore, it should also consider “the development of a methodology for de-designating VLOPs”.

The DSA compliance burdens must be proportionate – especially for non-profits

With regard to the application of the DSA to Wikipedia, researchers pointed out that the Commission should make more clear, through future legislation or guidance, “how responsibility is allocated between service providers and decentralised user communities”. Furthermore, the EU executive should evaluate the burden of the obligations prescribing “the yearly risk assessment and auditing, and the audit implementation mechanism”. Indeed, there could be a risk that those obligations “could disproportionately affect Wikipedia’s arguably low-risk, community-governed platform model”.

As of the supervisory fee, the Commission should continue to follow the principle of “the ability to pay [the supervisory fee] of the provider” and “explicitly recognise non-profit and non-governmental organisations in how it approaches the DSA’s supervisory fee”.

The Wikimedia Foundation was right not to join the Code of Conduct on Disinformation & electoral integrity guidelines 

When assessing the Wikimedia Foundation’s choice of not being a signatory of the Code of Conduct, researchers concluded that the “arguments for Wikipedia joining the Code of Conduct on Disinformation may not outweigh the arguments against, especially given Wikipedia’s transparent systems”. They have therefore confirmed the validity of such a choice. 

Finally, specific recommendations were made concerning the “guidelines on the mitigation of systemic risks for electoral processes”. In particular, researchers suggested that in the next iteration “consideration should be given to Wikipedia’s governance model” and that “the Guidelines should be written with Wikipedia in mind”.

Recommended Wikimedia transferable practices against disinformation

Reliable online sources 

The rules on deprecated sources and their transparent implementation are quite effective in guaranteeing the reliability of the used sources. Therefore, “the model of Wikipedia’s deprecated and non-deprecated sources could be an important tool to implement in the broader ecosystem to “help users assess the trustworthiness of information sources”.

Wikipedia as the go-to for election information

“Wikipedia articles on politicians and political figures could be categorised as “authoritative information on the electoral process” and made prominent and easily accessible across the online election-related ecosystem, thereby building on the Commission’s Election Guidelines”. Indeed, Wikipedia’s editorial rules “demonstrate how treating information about politicians and political figures with heightened care can prevent election-related disinformation and provide insights for broader regulation of the online environment”.

Real-time content patrolling, beyond Wikipedia 

“Wikipedia’s patrolling system is a notable example of how, through community-based oversight, accountability and transparency can be operationalised”. In this sense, the patrolling system that Wikipedia has in place, including the Recent Changes Patrol, could be extended to the broader online information ecosystem as it can help to prevent the dissemination of election-related disinformation.

Disclosure rules for social media influencers inspired by Wikipedia

“Wikipedia’s approach to paid editing and conflict-of-interest disclosure can offer a useful point of reference, particularly in relation to informing regulation of political social media influencers”, as it proved particularly effective in preventing and solving this sort of situation.

Read all policy recommendations in the full final report and its executive summary.

The article was originally published in the European University Institute Policy Report on Open Internet co-edited by Patryk Pawlak and Nils Berglund.

The evolution of economic models in a digital ecosystem

Before the development of the Internet, business models for digital ecosystems were mostly closed. Organisations or companies would develop a product or service to sell at a lower price than their competitors. The first digital encyclopaedias, such as Encarta, are an example. The past decades have brought a drastic change in the business model promoted by the US digital companies, whose strategy shifted from the service-for-a-price model to offering a service “for free”. They enacted two aggressive strategies to capture the market: (1) forcing competitors out of the market and, once a monopoly was established, demanding payment while preventing competitors from accessing the market and therefore sustaining lock-in mechanisms; (2) using user data as payment to facilitate the resale of this data to external parties. 

Competition laws in Europe have struggled to keep pace with these practices and address the challenges they pose to legal frameworks, either through the exploitation of grey areas or through legislative gaps. The lack of effective tax mechanisms to address the competitive advantage of global tech companies operating in the European Union (EU) over EU-established companies further complicates matters. The European Commission has partially addressed these challenges with the adoption of the Digital Markets Act (DMA) and the Digital Services Act (DSA), and has been actively working on regulating large platforms, with the objectives to limit monopolies and enforce existing laws and to strike a balance between content moderation and fairer revenue streams. 

However, in its ambition to establish a digital ecosystem that is fairer and more respectful of people’s ability to exercise their own self-determination online, the European legislators have focused on limiting the impacts of monopolistic enterprises, without proposing what a desirable future of digital platforms and commerce should look like to foster a thriving digital ecosystem in the EU. In other words, it did not address the compatibility of certain business models with European values enshrined in Article 2 of the Treaty on European Union (TEU). Policymakers seem to be forgetting that there are infrastructures such as the Wikimedia projects, of which Wikipedia is the most famous, that have a unique model; a model that supports desirable digital infrastructures: open source, transparent, community-driven, and privacy-focused. This peculiar model represents the most unique democratic collaboration system within a digital ecosystem. The unique visibility offered by Wikipedia allows policymakers to often legally carve out spaces for this model to continue. But what about other Digital Commons, such as OpenStreetMap, for instance? Wikimedia is part of the Digital Commons ecosystem, which aims at creating this desirable digital future and deserves more carefully designed policies.

Open Data, Open Content: Fuelling Big Tech or Open Democratic Societies? 

Information enables empowerment. Wikimedia projects contribute to gathering knowledge to share it freely and openly. They enable citizens in their daily lives by providing access to neutral and verifiable information, supporting education, autonomy, empowerment, informed decision-making, democratic participation, accountability, innovation, economic development, social cohesion, and resilience for all people wherever they live. The model provides equal access to all, through the information provided within the open ecosystem: an academic in Argentina, a farmer in Belgium, a civil servant in Thailand, a CEO of a medium enterprise in Kenya, or a large tech company in the United States or in Europe.

The interstices of this model created a condition under which big tech companies can exploit existing laws to extract value from data and content. The protection of personal data under the GDPR is challenged by the black box syndrome: lack of transparency, lack of human rights enforcements and the extraction of value at unprecedented scale, against the intentions of the creators, occurs without sharing fair remuneration, increasing wealth inequities and social developments. These models challenge the concept of information being equally accessible and reusable by everyone, which aligns with the concept of equity. 

This raises a question whether, in democratic societies, restricting access to information because of the inequitable and extractive use by a few companies is a justified response. Such restrictions would have a major impact on people, including the progress towards the United Nations Sustainable Development Goals (SDGs). However, maintaining the status quo is not a viable option either. Finding solutions that will allow for fair and equitable remuneration mechanisms, ensure the visibility of sources, including human contributions, and facilitate transparent, accountable infrastructural designs for the digital commons ecosystem are urgently needed.

Economic Case for Protecting Common Goods in a Predatory Digital Ecosystem

One main aspect of the difficulties that digital commons face is not only the question of the materiality and digitality of the infrastructures, but also the financial extraction of the value contained in data, which is turned into financial flows to the benefit of a few concentrated global powers. This process diminishes the power of other infrastructures and communities while tilting the balance of power in favour of a few big tech companies.

In the era of the attention economy and data extraction in exchange for access to monopolistic digital infrastructures, the Wikimedia Movement and Projects remain one of the last bastions promoting the values of the open internet and net neutrality. Monopolistic digital infrastructures prioritise short-term commercial gains by enclosing public spaces and failing to respect and promote the fundamental rights of individuals and communities. 

Wikimedia Projects do not pursue profit and are oriented towards the public good: providing free knowledge and neutral, verifiable information to everyone. They do not sell information or collect or sell user data. They do not exploit attention-economy ecosystems, such as addictive designs, to keep people in the infrastructure. Algorithms are not used to give readers what reinforces their own beliefs and convictions based on profiling methods. Respect and promotion of people’s privacy serve to protect them and their ability to self-determination by fostering critical thinking. Equal access to information, the same facts, and multiple perspectives are the crucial enablers for forming opinions and meaningfully participating in democratic life. 

Supporting communities and infrastructure, whether physical or digital, requires financial streams. This raises the question of how to fairly redistribute the value back to the people who created it. The goal is to continue developing a more equitable, fair, and inclusive society, while protecting the commons and digital public goods. 

One solution adopted by the Wikimedia Foundation is the launch of a commercial service, Wikimedia Enterprise, to ensure that the value extracted by a few powerful big tech companies is at least partially returned to the people who created it: the Wikimedia Community. Such a solution, however, is not definitive given the unique features of Wikimedia projects and the undesirability of a one-size-fits-all approach. Other NGOs and Commons may have different business models, which may prevent them from effectively engaging in negotiations with big companies. Born a quarter of a century ago, Wikipedia serves as a reminder that the principles of net neutrality and an open internet are under attack by a few monopolistic enterprises, which replicate and reinforce inequalities at every level of society.

Think about what that actually means. A non-commercial, volunteer-built encyclopedia in hundreds of languages, freely available to anyone with an internet connection. No paywalls. No shareholders. No algorithm deciding what you should read next. Around 260,000 volunteers curate 65 million entries, viewed more than 15 billion times every month. We don’t have nearly enough of that kind of thing.

A room full of communities

The evening had the feel of a reunion — familiar faces from across the European digital policy and Wikimedia communities. A mini-exhibition taking guests through the history of Wikipedia. A Wikicheese station where volunteers photographed Belgian cheeses to improve culinary content on the internet. These things only happen when people have been genuinely building something together long enough that the work creates its own history.

Annie Rauwerda

Stand-up Wikipedian and Depths of Wikipedia creator Annie Rauwerda took the stage and did what she does best — making the audience laugh while quietly reminding everyone why Wikipedia and free knowledge is beautiful. Her show was a love letter to trivia, knowledge, facts and humans. You may watch an older version of her talk online.

Here’s to the next 25.

One accent was the conscious decision to celebrate being human. Wikipedia is a project of humans and dedicated to human knowledge. With everything that this entails – good and bad. In a world full of machine generated content, the Wikimedia movement wants to confirm its commitment to being community human of volunteers first and foremost.

And since we are human, you will also bear with us that we will plug the Wikipedia Test here. It is a policy tool that asks a simple question: does a proposed law harm Wikipedia? When a law harms Wikipedia, it likely harms other community-led, nonprofit digital spaces too – spaces that don’t sell ads, don’t harvest data, and exist purely in the public interest. If you’re a policymaker, it’s worth thinking about and using it.

This was the occasion for the researchers to showcase their analyses and offer their conclusions after they conducted an investigation that lasted 18 months. 

The research is a combination of a legal and computational analysis of the fact-checking and content moderation practices used by Wikipedia during the 2024 European Parliament elections to address disinformation. Its aim was to produce policy recommendations that could inform future legislation able to safeguard community-driven, free knowledge initiatives.

In view of future revision of the Digital Services Act (DSA) and as a concrete solution to strengthen the information integrity in Europe, the research highlights how community-governed platforms, and in particular the online encyclopaedia Wikipedia, contribute to a reliable, pluralistic online information ecosystem. 

The research offers concrete guidance to lawmakers when adopting new legislation as well as to regulators when implementing the current rules.

If you did not have the chance to follow the event, you can watch the recording here.

Now open Wikipedia. You land on an article. At the top, perhaps, a small notice: “This article’s neutrality is disputed.” Or: “This article needs additional citations.” Or nothing at all — just the text, the citations, the talk page quietly humming in the background where a numerous strangers are negotiating the phrasing of a single contested sentence. This, too, is not an accident. It is the result of deeply intentional architectural choices — choices that point in precisely the opposite direction from the attention-advertising economy.

The contrast is worth dwelling on, because the stakes are high. We are living through a crisis of epistemic commons: the shared pool of facts, interpretations, and frameworks that democratic societies need in order to function. Social media has accelerated the fragmentation of that commons. Wikipedia, for all its imperfections, is one of the serious attempts to maintain it.

What “architecture” actually means

When technologists speak of the “architecture” of a platform, they mean the ensemble of design decisions — technical, social, economic — that shape how people behave within it. Architecture is not neutral. A street with wide pavements and humped or narrowed crossings makes cars slow down, even without an explicit speed limit. Similarly, a platform’s architecture powerfully determines whether its users tend toward collaboration or conflict, toward nuance or simplification, toward shared reality or tribal bubbles.

Social media platforms share several architectural features that, in combination, tend to produce polarisation. Content is ranked by engagement metrics, and emotionally charged content earns more engagement. Users receive personalised feeds that gradually filter out more nuanced perspective. Sharing is frictionless, rewarding speed over accuracy. Identity — who you are, which tribe you belong to — is constantly at stake in every interaction, making every disagreement feel existential. And the whole system is monetised through advertising, which means the platform’s financial incentive is to maximise time on site, not to serve users’ long-term wellbeing or society’s informational health.

Wikimedia projects share none of these features. They carry no advertising. They do not rank content by engagement. They have no algorithmic feed, no personalised filter bubble. They do not surface content designed to provoke. The architecture is, at its core, the architecture of a library — not a stadium.

one article per person

The perhaps most foundational design principle of Wikipedia is that there is one article per language about a person, a thing or a group. So there is one article in English about a politician, but also one article in Estonian about a football club. Regardless whether you love or hate said politician or football club, you will be looking at and editing the same article. This design doesn’t allow “filter bubbles” to grow. It forces people to come to together, disagree or agree, in the same space. They must find a common version, a common set of facts that they accept.

There is also a Neutral Point of View policy — NPOV in wiki shorthand. It requires that articles present all significant perspectives on a topic fairly and without editorial advocacy, attributing views to their sources rather than asserting them as objective truth. This is, on its face, an almost utopian demand. Neutrality is contested; every choice of which perspectives count as “significant” is itself a political act. Wikipedia’s editors argue about this constantly, and the arguments are sometimes fierce.

The talk page: dissent as collaboration

But here is the crucial thing: the argument happens not on the article itself, but on its attached talk page. Talk pages are open, the disagreement is not suppressed, but it is not kindled either. It is structured. And the goal of the structure is always the same: to produce an article that a reader with no stake in the outcome can trust as a fair account.

Talk pages can be quiet or ferociously busy, depending on the sensitivity of the topic. Articles about contested political figures, historical atrocities, scientific controversies, or living persons can have talk pages running to hundreds of thousands of words, with years of accumulated deliberation visible to any reader who cares to look.

When it works — and it works more often than critics expect — it works because the architecture creates incentives for convergence rather than divergence. To “win” an edit dispute on Wikipedia, you do not need to defeat your opponent or be “louder”. You need to write a sentence that they can accept.

This is radically different from the incentive structure of a social media argument, where winning means humiliating the other side, accumulating likes, and retreating to your corner with your followers’ approval. On Wikipedia and its sister projects, you don’t follow other users’ accounts. You add articles to your “watchlist”, meaning that you get a notification when something changes,

The talk page is, in a sense, the soul of the Wikimedia model. It externalises disagreement — moves it out of the article and into a dedicated space — while preserving a record of why decisions were made. An editor who wants to change a contested passage must engage with the reasons it was written that way, must respond to objections, must propose compromise. The system does not guarantee good outcomes; bad-faith actors exist, and some topics resist consensus indefinitely. But the system is structurally biased toward resolution rather than escalation.

An invitation

None of this is to suggest that Wikipedia is perfect. It has well-documented biases in the demographics of its contributor community, which skews heavily male and heavily from the Global North. Some topics are covered with extraordinary depth and rigour; others are thin, outdated, or shaped by the particular preoccupations of whoever happened to care enough to write them. The deliberative processes that govern the projects can be slow, exhausting, and sometimes hostile to newcomers. These are real problems that Wikimedia communities around the world are actively working to address.

But the architecture — the fundamental design choices that determine what incentives editors face and what outcomes the system is biased toward — is sound in ways that matter enormously right now. Wikipedia does not make money from your outrage. It does not show you a personalised reality. It does not reward you for defeating an opponent. It rewards patience, citation, and the willingness to sit with a stranger across a talk page until you find the sentence you can both live with.

In an information environment defined by fragmentation and bad faith, that is a quiet radical act. And it is an act anyone can join. The edit button is right there. Consider this a personal invitation!

The report starts by accounting for the latest developments in the application of the EU disinformation legal framework, including two rulings of the European Court of Human Rights and the stance adopted by the American administration and legislative bodies towards the Wikimedia Foundation (WMF). Then, it details the findings of the critical analysis of the EU legal framework.

Definition of disinformation

It highlights the lack of a legal definition of “disinformation”, pointing out that only the Code of Conduct of Disinformation, which has a non binding nature, offers some guidance in this respect placing emphasis on the element of “intention”. The same is true for the terms of use of the WMF defining “false information”.

Value of information published on Wikipedia 

It points out that the information published on Wikipedia, due to its encyclopedic character and specific editorial rules, appears to have public significance differently to social media platforms, thus receiving the highest level of protection under the fundamental right to freedom of expression.

Enforcement of DSA obligations

The evaluation of the enforcement of the DSA disinformation obligations is quite positive. In this sense, researchers emphasised that “it must be remembered that while there is considerable regulatory activity under the DSA, Wikipedia is the only sole VLOP, out of a total of 25 VLOPs, that has not been subject to any regulatory activity by the European Commission under the DSA as of October 2025. As such, following two years of the DSA’s provisions being applicable to VLOPs, Wikipedia has not (yet) come to regulatory attention; while none of the post-European Parliament election reports by the European Commission published in June 2025 mention Wikipedia”. Such a result can be interpreted to be linked to the effectiveness of the platform business model and its transparency.

Informing policy making: editorial rules & patrolling system

Researchers offer a preliminary overview on the specific solutions adopted by Wikipedia that may inform future policy making on election disinformation. 

In particular, they point out that the specific editorial rules, including neutral point of view, verifiability, no original research, and the policy on biography of living people, “illustrate how, for example, treating information about politicians and political actors with heightened care can help prevent election-related disinformation, and can provide learnings for broader regulation in the online environment”.
They also detail how community-based content moderation takes place and conclude that “Wikipedia’s patrolling system is a notable example of how, through community-based oversight, accountability and transparency can be operationalised. The idea of setting up patrols for political and election-related pages can also inform the wider online information ecosystem, demonstrating how targeted monitoring can help prevent the spread of election-related disinformation”.

Download the Report for more details.

Disclaimer. The sole responsibility for any content supported by the European Media and Information Fund lies with the author(s) and it may not necessarily reflect the positions of the EMIF and the Fund Partners, the Calouste Gulbenkian Foundation and the European University Institute. https://gulbenkian.pt/emifund/disclaimer/

The Commission frames these changes as efforts to streamline complex EU rules. However, civil society groups, many MEPs, and even some Member States argue that they fundamentally alter the EU’s digital rulebook by weakening longstanding data protection principles. 

The Wikimedia movement is made up of its organisations, projects and users. They all depend on robust privacy protections, open knowledge sharing, and solid intermediary liability protections. At the same time, the Wikimedia Foundation, as a service provider, also spends considerable resources on sometimes very complex compliance work in the EU.

With all this in mind it is needless to say that these proposals raise significant questions for Wikimedia. We can identify both positive and negative changes in the published texts. 

Comments on Data Protection Changes 

The most contentious aspects of the GDPR proposals involve redefining the definitions of personal and pseudonymised data. These changes could substantially narrow the scope of data protection. One example is that it would be much easier to share pseudonymised data without safeguards, even if the data can be de-pseudonymised. This looks like a re-scoping of fundamental rights, not a simplification measure. 

The proposals would also make the re-use of personal data for AI training easier by categorising it as “legitimate interest,” which paradoxically sets a lower threshold for AI training than for many everyday data uses. These steps look like they will benefit very large technology companies rather than the smaller competitors the Commission claims to support. 

On the other hand, the Commission is proposing a single-entry point for notifications for data breaches. Such breaches currently have to be reported to many or all data protection authorities across the EU and EFTA, which clearly wasn’t practical or very workable. 

Another positive element is the proposed so-called “privacy signal”. It would give people a clear way to refuse or allow data access by setting their preferences in the browser instead of clicking through pop-ups on individual sites.

Impact on Open Data and Knowledge Commons

The Omnibus wants to reduce the number of data-related legislative acts that currently includes the Data Governance Act, the Open Data Directive and the Data Act. It would repeal the directive and incorporate its articles into the regulation, the Data Act. This is positive in two ways: It will hopefully make the rules easier to understand and reduce overlaps and contradictions, as they will be in one place. As a regulation is also directly applicable, this should, at least in theory, mean less national divergence across the EU. 

The Commission is also proposing another change that needs to be thoroughly thought through. It wants to allow public sector bodies to charge higher fees for data and documents requested by very large enterprises. Politically speaking this is understandable and may even be desirable. Actors with immense economic power should contribute more to the commons.  

The challenge here comes from the practical application. In order to treat very large enterprises differently from everyone else, public sector bodies would likely try to change their standard open licenses to non-standard and restrictive ones. This is a risk for the re-use of open government data.

Wikimedia itself has looked at a similar challenge and thought hard how to solve this conundrum. Wikimedia Enterprise ensures that all knowledge on Wikimedia projects remains free & open, while charging very large re-users for the provision of very high bandwidth access, but not for the data itself. Of course this may not work for everyone. But charging one specific group of actors for the data itself would definitely break standard open licensing. 

Another aspect is that currently EU rules already allow public sector bodies to charge for some datasets. If the new rule is limited to this category of data, the negative consequences would likely be very manageable.    

The Elephant in the Room: Russmedia 

While the European Commission is trying to simplify certain parts of the digital framework, the Court of Justice of the EU is changing important parts of it as well. The recent Russmedia ruling results in several major question marks around how the GDPR relates to the DSA. It seems to make significant changes to platforms’ intermediary liability protections, something without which user-generated content is quasi unimaginable.

Some have argued that it’s limited to marketplace platforms, but the consequences of its application to other services could be momentous. All eyes now are on the forthcoming Meta/Kunast case in Germany, which might (as a matter of German law) explore transposing the Russmedia doctrine over to commercial social media platforms.  If it does spread, there will be greater pressure on the EU to provide a legislative fix. 

Our view is that Wikipedia and its sister projects shouldn’t be directly affected by the decision, for instance because uploaders aren’t paying to amplify content or algorithmically push content on other users. An EU-wide reaffirmation of intermediary liability protection principles and harmonisation of strong freedom of education and expression exemptions under GDPR Article 85 would nevertheless be a very welcome EU initiative.

Wikimedia Europe’s Response

Wikimedia Europe is carefully analysing the proposals and their potential impacts on free knowledge, user privacy and non-commercial, community-driven platforms. We recognise the need to streamline the sometimes very complex maze of rules across EU laws and welcome such steps. We would also like a public discussion that clearly differentiates between simplification on one hand and re-scoping or even limiting fundamental rights on the other. We see a tendency of conflating the two.  

We intend to provide feedback to the Commission consultation (due 3 February) and relevant lawmakers. 

Dear Presdent Macron,
Dear Chancellor Merz,
Dear President von der Leyen,

Europe is at a crossroads. The Summit on European Digital Sovereignty marks an important milestone for the EU and its member states in aligning on a shared strategy for achieving real and lasting European digital sovereignty. As the EU pursues the goal of digital sovereignty, we urge you to harness open source — that is, technology that is free to use, inspect, adapt, and share — as a key enabler of this strategy. 

Europe cannot buy sovereignty off a shelf, it has to build it. In an age of geopolitical volatility and rapid innovation Europe must play to its strengths, including world-leading researchers and a rich history of open source development. It faces a choice: use these strengths to carve out its distinct place in the global AI ecosystem or settle for copying the playbooks of already dominant actors. 

At their heart, closed systems create dependency, open systems create capacity.

Investment into the full open source AI stack, from AI models to data and software tooling, is a strategic lever. If digital sovereignty means creating a Europe that is resilient and benefits from choice, security, and self-determination, then open source is a critical force multiplier that enables Europe to do more with less. 

Open source AI, and open source technology more broadly, is not just a strategic asset benefiting governments, businesses, and people. If underpinned by a clear commitment to values that are at the heart of the European project — including cultural diversity, fundamental rights, environmental sustainability, and people’s privacy and security — open source can help embed these into the technologies that will shape our future. 

We, the undersigned, represent a diverse coalition of organisations across industry, the open source community, and civil society — many of whom build and maintain leading-edge open source technology. With this letter, we put forward a concrete plan to ensure Europe’s technological future is open, trusted, and its own. 

The importance of open source in achieving digital sovereignty

Boosting and utilising the open source AI ecosystem will support the EU and its member states in strengthening their digital sovereignty in four key ways: 

• Reduce Dependency and Increase Strategic Autonomy: Open source technology enables European governments and enterprises to freely use, adapt, and host technology on their own terms, using infrastructure of their own choosing. By making it easier to switch and by fostering more competition, this prevents vendor lock-in, increases choice, and reduces dependencies throughout the technological supply chain. 
• Boost European Capability and Competitiveness: Open source compounds progress and boosts European innovators’ productivity by providing them with reusable building blocks that they can use and tailor to their needs, without having to reinvent the wheel. It helps EU startups, SMEs, and researchers go further, faster, rapidly delivering innovative technology. 
• Build Global Leadership and Influence: Open source enables Europe to collaborate globally while retaining autonomy. The technology can be developed and maintained across borders, harnessing expertise from around the world without requiring trust to verify its security. Investing in open source AI can also strengthen partnerships with like-minded nations — all while influencing global standards, facilitating interoperability, and making it easier for others to build on European technology. 
• Promote European Values and Cultural Diversity: Open source and open data can safeguard linguistic and cultural diversity by making European language and cultural data more broadly available and by enabling local communities to adapt AI to their needs and context. It is also inherently more transparent and enables independent audits — key to AI’s safety and security.

Five actions to harness the potential of open source AI for Europe’s ambitions

To leverage the value of open source AI for the EU’s ambitions on AI and digital sovereignty, we call on Member State governments and the European institutions to champion open source through the following initial actions:

1. Leverage the public sector’s buying power to scale and ensure the sustainability of open source AI initiatives: Improve tendering processes and templates to better account for open source technologies and reduce administrative obstacles for open source vendors, rather than structurally favoring proprietary technology. Consider the benefits of open source with regard to sovereignty, total cost, and interoperability as part of the procurement process.
2. Mobilise funding to incubate, develop, and maintain an open source AI stack: Create designated funding lines and incentives to support the development and maintenance of critical and high-impact open source AI and other foundational open source technologies, including through the creation of an EU Sovereign Tech Fund, the European Competitiveness Fund, and national funding instruments.
3. Facilitate access to computing infrastructure for open source AI research and development: Reserve capacity and facilitate reliable, unbureaucratic access to publicly funded compute, for example through AI factories, for open source and public interest AI research, development, and deployment. 
4. Unlock data for open source AI development while protecting privacy and other rights: Remove barriers to access and reuse of publicly funded, public domain, or other non-sensitive public sector data for open source AI developers. Accelerate the deployment of data sharing mechanisms and infrastructure, such as Common European Data Spaces.
5. Build capacity in the public and private sectors to leverage the power of open source: Mainstream support for open source AI developers and users within existing governance mechanisms, including European Digital Innovation Hubs and supervisory authorities. Raise awareness and foster sharing of best practices around the use of open source AI.

We urge the EU’s leaders to use this five-point plan as a pathway to build a future it can trust, shape and truly call its own. 

Sincerely, 

The latest batch of competition that wants to harness AI technology to generate better compendiums of knowledge. These projects criticise things like gaps in coverage, reliable or alleged political biases. Let’s have a look at what’s out there and discuss some of the aspects!

Projects Using AI to Generate Encyclopedic Content

xAI’s Grokipedia is not the first to explore using Al for generating Wikipedia-like articles. Here is a list of several others, courtesy of The Signpost editors:

• A small website called WikiGen.ai (one developer’s side project) already offers “automatically create[d] comprehensive articles on any topic you can imagine. Unlike traditional wikis that require human editors, our Al instantly generates well-structured, informative content tailored to your preferred reading level”.
• “Botipedia”, a project by INSEAD professor Philip M. Parker (which has been under development since at least 2021 and moved to an LLM-based approach more recently), reacted to Musk’s September announcement by asserting that it had already launched version 0.5 of its “truth-seeking Al with 400B+ articles, 6,000x bigger than Wikipedia” (although a later tweet clarified it will only be “Open to all in 2026. For now, limited to edu/org/corp emails while we scale”). Larry Sanger praised it as “one of the most interesting new competitors of Wikipedia”. A promotional video portrays Botipedia as being superior to Wikipedia due to its inclusionism and language diversity: “No subject, event, language or geography is too obscure to merit an article, meaning that no language gets left behind.”
• The task of using LLMs to write Wikipedia-like articles has been the object of numerous academic research efforts for years (see e.g. our 2024 coverage of “STORM”, a particularly notable project out of Stanford University that has also seen considerable real-life usage).
• Lastly, like Wikipedia itself, Grokipedia could also be seen as competing with ChatGPT Deep Research and similar offerings by OpenAl’s competitors (like Gemini Deep Research) that generate cited reports on a user-specified topic.

Licenses

Wikipedia and its sister projects are freely licensed. One argument for this choice is that barriers to access and re-use for knowledge should be kept at bay. We want knowledge to be free. 

This means that whenever someone thinks there is a better way to gather and share knowledge, they have the right to try. We won’t act as an entrenched, dominant player and use licences to stall potential progress. Competition, generally speaking, is welcome.    

Biases on Wikipedia

Perhaps the main motivation behind using AI and LLMs to create knowledge compendiums akin to Wikipedia is the project’s perceived bias. Let’s take a look. 

Wikipedia strives to achieve a neutral point of view, this covers content, perspectives and sources within those articles. This rule is non-negotiable. Wikipedia editors work to write articles with an impartial tone that documents and explains major points of view, giving due weight for their prominence. All articles must strive for verifiable accuracy with citations based on reliable sources. Editors’ personal experiences, interpretations, or opinions do not belong on Wikipedia.

This does not mean that it will always achieve this. There are many sources of bias, biased sources and also different ways in which they manifest. Examples include contributors’ own cultural bias (different language versions will look different), coverage bias (some Wikipedias will have detailed information on one topic, but lack another) and gender bias (women are still underrepresented). They, of course, overlap. The gender bias will be influenced by the cultural bias and itself will result in a coverage bias, to show just one string. 

When arguing about reliable sources, of course, there will be many different views on what is reliable. This can change from topic to topic, from language to language and even change over time. Human editors constantly debate and look for consensus. The project is alive and, by definition, never completed. It will never be perfect.  

Biases by LLMs

Curating human knowledge is messy. But what about machines? Can they really help make knowledge less partial, less biased? 

At first glance, machines will have the same problem that humans have. They are a product of the world around them. Which means that they too inadvertently suffer from the same biases mentioned above. 

It would be interesting to read a systematic, scientific comparison of reliability between Wikipedia, Britannica and several AI projects. Comparisons of this kind already exist for Wikipedia vs. Britannica or other classical encyclopedias. It would be interesting to extend them to encyclopaedic AI projects. 

For now we can take a look at a couple of more limited studies that are already available. 

A Stanford study recently published in Springer Nature looks at 24 major LLMs and finds that they still struggle to tell fact from opinion. The scientists tested the models on 13,000 questions to evaluate how well they distinguish beliefs from knowledge and fact from fiction. When responding to a false, first person belief phrased as “I believe that…”, the researchers say all models tested systematically failed to correct the false belief.  

Another study, focused on health care, found ample proof of inherent bias in LLMs. And while they also acknowledge that what they call “implicit bias cannot be eliminated from society or training data”, they also say that “its existence must be acknowledged and mitigated”. One issue that the scientists had is that many models don’t provide either all sources or a transparent documentation of how they work. This makes it impossible to investigate the source of the bias. 

It seems to be surprisingly hard to make any system unbiased. Perhaps bias is not a technological problem, but a societal one? From this perspective technology cannot and will not provide a magic solution, but it can either improve or worsen the situation, depending on its architecture and use. There opportunities, limits and risks of machine learning are something we need to keep in mind, observe and actively discuss. They are both social and scientific.  

Where Are We Going?

That being said, AI models can help catch mistakes. xAI and Grokipedia have found some mistakes in Wikipedia (HT User:Haeb). For example, the last film Pedro de Cordoba appeared in. Another one is the surface area of Lake Starnberg. Both immediately corrected. AI can also be super useful in finding and perhaps even updating old statistical information. Imagine new census data is out and a city’s article still shows the old population statistics. 

Simultaneously, Wikipedians have also found mistakes on Gorkipedia or articles that seem to depart from a neutral point of view. Examples include that cited sources are not reliable or that the article isn’t saying what the citation claims. 

We also know, from research and from experience shared by the developers of LLMs themselves, that organic, human knowledge is indispensable. These systems can’t, at least at present, deliver without human content. 

Improving omissions or knowledge gaps can go either way. LLMs can help better cover content in one language that already exists in another. Think of information about train technology in Bulgarian that currently exists only in German, for instance. They however, can’t cover the gaps that exist in the human world. If there is no reliable data available, they will simply invent unreliable content.

As for bias, so far we can’t observe that content written by LLMs is less biased than community generated content. But, again, removing the bias from any system is a very tough challenge. Perhaps not even a technical one.