Introduction

Attribution has always been a cornerstone of working with other people’s creativity and knowledge.

As Creative Commons reminds us, practicing attribution serves many important functions, from providing verifiable evidence of a claim, to paying respect to the work of others, and creating pathways for traffic and financial value to flow.

Despite the importance of attribution, many of today’s AI systems fail to acknowledge the sources of knowledge and creativity that make them possible.

A recent study led by the AI Disclosures Project found that more than 30% of responses by leading search-enabled Large Language Models (LLMs) provided no attribution whatsoever. Even where attribution is provided, it can be limited in ways that are hidden to users, and deep technical challenges persist in connecting outputs from AI to the sources they are derived from.

Despite this, we believe there are good reasons to be optimistic that AI systems can attribute the sources they use. Not perfectly, and not always, but enough and improving, such that we should reject the argument that a lack of attribution is an inevitable side effect of the way the technology works. The commons needn’t become “hidden substrate.”

A Note on Terminology
In this post, we interpret attribution as the practice of connecting an artifact with the sources of data used to create it. This is deliberately broad. As we explain, it’s possible—and useful—for AI systems to practice different forms of attribution for different purposes. We are not referring to attribution as a specific legal requirement or condition. We also use “AI systems” broadly, as most of what we discuss applies to a wide range of models and surrounding technologies (although some methods apply mainly to systems that generate text outputs).

Reasons to Be Optimistic

1. AI systems have evolved in ways that support attribution.

The first wave of LLM-based tools were essentially user interfaces that made calls to a static, pre-trained model (or “thin ChatGPT wrappers”). Outputs from these tools were a function of whatever the model had been exposed to in training, with no live access to sources.

This changed quickly. Now, architectures such as retrieval augmented generation (RAG) enable AI systems to draw on fresh data (such as web pages or a user’s files) to produce outputs.

This practice of retrieval makes AI systems much more amenable to attribution. When systems generate an output based on retrieved content, they can be trained and prompted to attach claims in the output to specific sources where those claims can be corroborated. This provides a form of credit, and allows users to follow through to web pages or trace back to their own files.

Claude, for example, can provide footnotes with clickable links to sources. To do this, it breaks the source materials it retrieves into “chunks”, each with a unique ID. Under default settings, it will—selectively, at least—cite the IDs of chunks that support the claims made in the output and render them as clickable links in the footnotes. Claude’s Citations feature can be used to make this practice more frequent and consistent, and enable outputs to point to exact passages from users’ files. It works by extracting source chunks, such as quotations, and injecting them, verbatim, into the output.

Other approaches exist. Pleias develops small, open-source AI models using openly licensed and public domain data. Pleias-RAG models are designed to produce outputs exclusively from retrieved sources. Rather than chunking and injecting, the model generates both the output and the supporting quote itself, in a single pass. This approach is more flexible, as the quote isn’t bound by pre-defined chunks, but it means that the system cannot fully guarantee that its quotations perfectly match the source. For a user, the result looks similar: an output whose claims trace to sources, with quoted excerpts and clickable links (if the application built on top of the open-source model makes this design choice).

There’s room for improvement here. Most AI systems do not attribute their retrieved sources as systematically as the Pleias-RAG models. Even where systems do, it often doesn’t amount to the full Title Author Source License (TASL) format that CC recommends. And frameworks like OpenMined’s Attribution-Based Control demonstrate how systems could practice a more decentralized, privacy-preserving form of retrieval, where data is queried where it lives rather than copied from the user’s system or the web into a central index.

In the round, retrieval architectures have enabled attribution practices that represent a strong improvement from the first LLM-enabled black boxes we were exposed to. These developments have led researchers from the AI Disclosures Project and their collaborators to conclude that “attribution gaps result from design choices, not simply technical limitations. The best-performing systems show that transparent, comprehensive attribution is technically feasible today”.

2. Data attribution is an active research discipline.

The approaches to attribution described above can provide corroboration, but they don’t truly establish that a listed source caused an output. They often list just one source, and only address the sources that a system accessed at inference time.

Using these methods, a system could rely on multiple sources to produce an output, across training and inference, but list only one source as corroboration.

Research is underway to address the harder question of: “given everything the system drew on, which sources shaped this output, such that removing them would have changed the result?”

Some methods attempt to answer this by reading the model’s internal state to estimate which inputs mattered to an output. Attention weights work by monitoring a model’s focus across retrieved source materials as it generates an output. The share of attention landing on a given source is sometimes used as a proxy for how much it mattered. Gradient-based methods instead measure sensitivity, seeking to calculate how much the answer would shift if each input were changed slightly, on the assumption that the inputs the output is most sensitive to are the ones that shaped it. 

A different cluster of methods follows the logic of “leave one out” to try to identify how an output would change if one of its source materials was removed.

Doing this perfectly would involve rerunning inference with different sources withheld each time. That gets computationally expensive quickly, as attributing at the level of individual passages could mean hundreds of reruns per output. Most research attempts to approximate this effect by using different methods to cut the cost of producing the counterfactual. AttriBoT approximates the change in a system’s response when particular spans of retrieved sources are taken away by testing large blocks of source content first before “zooming in” on smaller spans. ContextCite looks at random subsets of the sources to predict how the answer shifts as sources are switched on and off. OpenMined is experimenting with Shapley values, which work by averaging each source’s marginal contribution across many different combinations of the sources.

This same counterfactual logic can also be pointed at training data rather than retrieved sources.

Influence functions estimate, without retraining, how sensitive the model’s output is to each training source and, therefore, how much removing one would change the answer. Doing this involves inverting a matrix that records how every parameter in a model interacts with every other parameter. Its size is the full model’s parameter count squared. For today’s models, which often exceed 3 trillion parameters, this is far too large to even store, let alone compute with. As a result, most progress on training data attribution involves producing estimations for which training data point’s absence would have changed the model’s output the most. Researchers at Anthropic have experimented with a technique called EK-FAC to approximate that enormous matrix with a simpler, structured stand-in. Arnoldi iteration keeps only the matrix’s dominant directions. Others cut the cost differently. TracIn skips the matrix and instead observes the model parameters at different points in training to assess how a particular input shaped the output, and TRAK treats the trained model as if it was a simple linear one, which makes the same estimate far cheaper to compute.

Researchers have found that estimates of training data contributions can be fairly accurate for small, simple models, but become unreliable quickly as models grow, and different methods are likely to disagree on the impact of a particular source. Also, while they can—perhaps—point to the individual training examples that most shaped an output, those examples are unlikely to have information about their authorship or provenance attached. Identifying an influential training source is not the same as being able to properly attribute it.

While imperfect, the methods described here are the product of a growing field of research dedicated to data attribution, and it’s reasonable to expect some will work their way into mainstream AI systems. Taking an even wider view, most of the systems we have today were not built to preserve information about the sources they use, especially in training. We are hopeful that more source-aware design choices will be made from the ground up (such as more curated and persistently identifiable training data), rather than attribution needing to be bolted onto systems that chose not to anticipate its importance.

3. Regulatory pressure on AI developers to disclose their sources is increasing.

Left to themselves, AI developers have disclosed little about the content their models are trained on or have access to. Many seem to view training data disclosure, in particular, as an unnecessary burden, or its composition as a source of competitive advantage and legal exposure.

In response, regulators have started to introduce minimum standards for what AI developers must document and publish about the data used by their models.

The EU AI Act, for example, requires providers of general-purpose AI models to publish a “sufficiently detailed summary” of their training content, using a template published by the European Commission’s AI Office. The template requires an organised account of the model’s main data sources, including large publicly available datasets, data licensed from rightsholders, and data crawled or scraped from online sources. California’s Generative AI Training Data Transparency Act similarly requires developers to publish a summary of the datasets used to train any generative AI system made available to Californians. 

In this context, attribution involves connecting sources of training data to a model rather than to a specific output. It’s the difference between answering “what sources made this model capable of producing outputs?” vs “what sources has this model used to produce this output?” Answering this meets different needs to other forms of attribution described in this post, which are generally focused on connecting outputs from models to the sources used to generate them. Model-level attribution (often referred to as simply “transparency” or “disclosure”) enables consumers to compare competing systems, regulators to supervise and enforce regulation, and researchers to make high-order assessments of things like bias and safety.

This form of attribution is typically summarized rather than granular. Summaries might only individually name large—rather than all—publicly available datasets and other sources of training data. And it’s often practiced with carve-outs. The EU template, for instance, lets model providers redact some elements they can justify as trade secrets.

While they serve different purposes, we think it’s useful to see output-level and model-level attribution as related. First, the kind of source-aware design that model-level regulation increasingly encourages, such as the use of more curated training data, is also what makes better output-level attribution possible (as we discussed earlier in this post). Second, we expect regulators to evolve their demands of AI systems to include output-level attribution. In June 2026, the UK’s Competition and Markets Authority imposed a conduct requirement on Google, under which it must take steps to ensure its search content is attributed clearly and accurately, as well as publish user-friendly information explaining its attribution methods. (However, on that note, we should be wary of attribution regulations that are overly prescriptive, onerous, or impractical, especially for open source and non-commercial development of AI, or for machine use more broadly.)

4. Users expect attribution.

After the first generation of LLMs became notorious for “hallucinations”, a system that outputs confidently without documenting its sources is now seen as deficient.

Users have come to expect to be able to click through to source material behind a claim, and that expectation shapes what gets built. It is part of the reason why systems like Claude now return outputs with links back to the web pages and user files that have been drawn on. 

In regulated and sensitive markets, this expectation is hardening from a consumer preference into a procurement requirement. Buyers of AI systems in sectors such as law, medicine, and finance increasingly require auditable trails showing what a given output was based on, in order to satisfy their own compliance regimes. Designed for medical professionals, OpenEvidence attributes its sources through much the same approach as Claude, and further improves reliability by constraining those sources to vetted medical literature.

As individuals, enterprises, and governments push for “sovereign AI”, systems that behave as an opaque blender, indiscriminately mixing everything they can find, will not suffice. The architectural choices that make control of AI systems’ relationship with data possible—such as permissioned access, federated retrieval, and transparent retrieval logs—are the same ones that make for better attribution, especially at inference time.

There is pressure from the data supply side, too. While licensed data deals can be concerning to those of us with an interest in broadening the digital commons, they create commercial incentives and contractual obligations for AI firms to develop means of identifying which sources contributed to which outputs. ProRata uses attribution as the basis for licensing and payment (though their methods are unpublished, making it hard to tell what they are really doing or how accurate it is); Perplexity’s Comet Plus pays participating publishers when their content is visited, cited in an answer, or acted on by an agent; and Index by Parallel uses Shapley values to encourage writers to “see what [their] content is worth to AI agents.”

While traditional formats like TASL still have a role to play, especially at inference time, we should be open to new patterns and standards for presenting attribution information to users. For Shapley-enabled methods, OpenMined’s research shows how charts or other design features can present the relative contribution of several sources to a single synthesized output. The Wikimedia Attribution Framework describes how including reference counts, contributor activity and calls-to-action can support the credibility of AI-generated responses, and drive contributors and donations to sources. In scholarly settings, participants at a recent Cambridge workshop flagged the need for outputs to present information such as retractions and other expressions of concern.

Moving Forward

We must not lose sight of the fact that attribution is a vital element of the social contract of sharing. Regardless of how machines are involved, naming our sources sustains the connection between a piece of creativity or knowledge and the people who produced it.

While it is indeed technically challenging, we’re optimistic that modern AI systems can meaningfully practice attribution, especially if attribution is incorporated by design. As this post has described: retrieval architectures make corroborative attribution eminently possible; more advanced, contributive methods are inching from research toward production; and commercial, regulatory, and user pressure are aligning behind the same goal.

As attribution practices mature, it’s important to us that they extend to creators, publishers, and stewards of openly licensed and public domain content rather than become the preserve only of those who engage in restrictive licensing deals or use sophisticated technology to gate access to their content. 

Lastly, we shouldn’t lose sight of the fact that getting attribution right is necessary but insufficient for the future of the commons. A creator can be attributed and still lose the traffic that funded their work. A public library can still find itself overwhelmed by machines drawing on its collection en masse. A community like Wikipedia can still lose its connection with contributors, current and potential. To build a genuinely reciprocal relationship between AI and the commons, we also need bold new ideas, infrastructure, and forms of exchange to sustain the communities and institutions the commons depends on.

If you are working on attribution, especially building AI systems that practice it in some of the ways we describe here, CC would like to hear from you.

The post Attribution in the Age of AI appeared first on Creative Commons.

The Creative Commons (CC) Open Heritage Companion is a practitioner’s guide developed to help cultural heritage institutions navigate the journey from closed to open collections.

Grounded in the legal, policy, and ethical questions institutions face along the way, the Companion brings together 200+ curated resources, tools, and examples, organized into six chapters that guide practitioners from their first questions about openness to leading the field.

A community full of tools, scattered

Over the years, the open culture community has produced a remarkable amount of practical knowledge: guides, toolkits, case studies, policy templates, and more. What’s been missing isn’t expertise but rather a single place that points to all of it. The Open Heritage Companion started as an effort to close that gap, bringing the community’s scattered knowledge into one navigable guide.

A six-stage journey

The Companion is organized around six stages: Understanding, Choosing, Preparing, Sharing, Impacting, and Leading. This framework wasn’t designed first and filled in after. It emerged from the resources themselves. While grouping resources gathered for the Companion, a pattern revealing categories surfaced. Some resources spoke to understanding the case for opening, others to choosing a licensing approach, others to the practical work of preparing collections, and so on through sharing, impacting, and leading. The six stages make that pattern explicit and draw a map of the journey informed by existing resources.

A living resource

The Companion is available in English and French, aligning with UNESCO’s principal working languages, and Spanish, extending the Companion’s reach further.

These PDFs reflect the Companion as it existed at the time of publication in July 2026. We hope that this resource list will keep growing beyond it. To explore the Companion, propose an addition, or follow future updates, visit the CC Open Heritage Companion on Meta-Wiki.

Testing the Companion

Before release, the Companion went through two rounds of community input and consultation with the Open Culture Platform, in addition to CC staff review. We then tested the Companion live by incorporating it into a capacity-building “think and do” workshop held during Oulu Löyly, part of Oulu, Finland’s 2026 European Capital of Culture activities.

We know that a tool or guide is only as useful as the motivation to open it, understand it, and implement it. The Companion is a rich resource on its own, but knowing where an institution stands on the journey from closed to open makes it easier to know which resources to start with. That was one of the barriers the Oulu Löyly workshop group set out to address.

Working with the Companion’s data, the group created ORLA, the Open Resource & Learning Assistant, a chatbot self-assessment designed to help a practitioner identify where they sit in the six-stage journey and point them to where to start. From there, they can go straight into the Companion’s curated resources.

ORLA is an early-stage beta, built during the workshop to test the idea in practice, not a finished or officially supported tool. It’s meant to complement expert guidance, not replace it. Use it to orient yourself, then bring what you learn into conversation with colleagues, mentors, or the wider open heritage community. As an LLM-powered tool, ORLA can get things wrong, so treat its suggestions as a starting point rather than authoritative advice. Running it requires your own API key or a self-hosted model, and any input you provide goes to that model provider, not to Creative Commons.

The code is on GitHub. You can try it here, and the full workshop documentation is here.

A shared starting point

The Companion exists because so many people were willing to share what they learned. It combines the work of the many practitioners, institutions, and organizations that spent years advancing open heritage. We are deeply grateful to those pioneers of openness who made the path easier for those who followed. Thank you to the members of the Creative Commons Open Culture Platform (now organized as the Open Culture Community of Practice), whose expertise, feedback, and recommendations shaped this resource. We are also grateful that this work was made possible with the support of Arcadia, a charitable fund of Lisbet Rausing and Peter Baldwin. 

Finally, to those who will be inspired by this Companion to take the next steps towards openness, thank you in advance. Every institution that opens heritage brings us closer to a world where heritage is equally accessible as a shared resource and culture can be enjoyed as a global public good. 

The post Introducing the CC Open Heritage Companion appeared first on Creative Commons.

At the recent Founders Fireside Chat, Creative Commons founder Lawrence Lessig referred to himself as “hopelessly naive.” It struck me that this sort of naivete—the kind that is about believing in the potential for good when sharing knowledge with other humans—is the glue that holds the entire Creative Commons movement together, across borders and belief systems.

Today, it can feel difficult to find that kind of naivete, or hope. In the age of AI, we all necessarily feel small. A lifetime’s worth of creative output becomes a microscopic portion of data in a system functioning at machine scale. 

No wonder our avenues for agency feel so small. We seem to have two options: share in a free-for-all system, or close off. 

In this unsatisfactory binary, many are choosing to close off, even those who have historically believed in the power of sharing knowledge. We all stand to lose. 

Creative Commons wants to create an alternative that sits between the extremes. We think there should be standardized tools for sharing knowledge in ways that reflect the spirit and values of the open movement, while responding to the concerns that many have about unfettered machine access. This is the work we have been doing as part of the CC signals project. 

To advance this work, we have been regularly holding convenings and engaging with people at events around the world. We recently held a workshop in London, bringing together policy, legal, and technical experts to discuss the evolution of the CC signals framework into legally enforceable tooling. As with our broader community, the workshop participants held wide-ranging views on the use of open content for AI development, including opposing ones. 

A few clear takeaways emerged from the workshop.

There is clear momentum for innovation in legal tooling. 

While there were dissenters, overall there was a recurring theme that CC is well-positioned to engage in product development designed to help the open movement evolve to meet the needs of the AI environment. 

The time is now.

The window of opportunity to influence the AI ecosystem is short before new standards and norms become entrenched. 

We must minimize unintended consequences. 

If we are not careful as a movement, new tooling could go too far to meet the demands of creators in a way that does not result in a net positive for the commons. As CC has done from its inception, we will balance the need for the rights of the public to make use of knowledge with creator agency.

We must simplify and test. 

Given the complexity of the legal and technological environment, it is not always clear in advance what will have a meaningful impact. We need to experiment in small ways, rather than trying to solve multifaceted problems at once. 

The devil is in the details. 

Even among those who agree with the direction toward new tooling, opinions about the details about how the tools could operate vary widely. There are no easy answers, and every aspect of product development will take great care.

All of this has led us to refine a better, and more informed, path forward. We are simplifying our plans for CC signals, breaking them into their component parts as part of a research and development effort to create new tooling. Importantly, we are planning to develop this tooling with a small group of data stewards who will serve as pilot testers. This narrowing will allow us to act quickly with real-life scenarios, while minimizing potential collateral damage as we experiment. The long term goal continues to be a suite of tools available for general purpose use, and we believe this is the fastest responsible way to get there. 

Over the next two months, we will be engaging with and learning from potential pilot adopters. If you are a steward of a collection of open content that is used as data for AI development and are interested in being a pilot adopter, please reach out to us at legal@creativecommons.org. 

We want to thank those experts who volunteered their time to help us think through these issues – and more – at the workshop. Every varied viewpoint makes for a stronger roadmap. We couldn’t have made this happen without the generous support of Morrison Foerster, providing space, food, and, most importantly, intellectual firepower.

Join us for upcoming convenings. 

Wikimania 2026

• Tuesday, July 21: CC Salon in celebration of CC’s 25th anniversary.
• Thursday, July 23: CC Workshop “Sharing & Collective Action in the Age of AI.” 

CC Symposium

Save the date for a CC Symposium on Monday, September 21, ahead of the 81st session of the United Nations General Assembly in New York City. More details to come.

The post Takeaways from a London Workshop on CC Signals appeared first on Creative Commons.

We are thrilled to recap the first in our series of 25th anniversary events, celebrating some of the most notable moments in the history of the development and deployment of our legal tools from creation to translation; and from advocacy to adoption. Each event highlighted both the ways in which CC licenses created unique movements within each sector but also the commonalities between fields and the people who have dedicated their careers to better sharing.

The Legacy & History of Open Education

Our first panel of experts were from the field of Open Education: Kathryn Kure, Dr. Cable Green, and Dr. Angela DeBarger, moderated by CC’s Jennryn Wetzler. 

The Open Education panel explored how openness transformed learning by challenging traditional models of access and ownership. Open educational resources (OER) have enabled educators and learners around the world to adapt, translate, and build upon materials, creating more inclusive and locally relevant learning experiences. CC licenses are what made this possible at scale by giving users a standardized, machine-readable way to mark materials, which meant institutions didn’t have to negotiate permissions one by one. The shift from bespoke agreements to common infrastructure is what allowed OER to grow from isolated pilot projects into global practice.

All three panelists recognized the foundational role of policy in the open education movement. Dr. Green shared that whenever he thinks of open education, he brings it back to the very start. “For me  the foundational principle that everything is based on in our work is the UN Universal Declaration of Human Rights Article 26 that says the right to education is a fundamental human right. This is our right—for everyone.” Dr. DeBarger echoed this sentiment in speaking about UNESCO’s Recommendation on OER. “[It] was certainly a catalyst in bringing visibility and validity to the work.”

However, speak to any educator and you will know that while policy and theory matter, it is the implementation, the on-the-ground practical work to bring that policy to the classroom that truly makes a difference. Kathryn Kure shared, “I like to see theory become praxis. In South Africa, many of the problems we’ve been looking at solving are challenges of a place with low to intermittent electricity, no data or limited data, a Moodle box on a Raspberry Pi. Incredibly under-resourced. And there’s this huge disparity between the Global North and Global South, where you can have a cornucopia of data or content here, and the reality on the ground is very different.…You understand that there are real challenges in implementing it.”

Dr. DeBarger summed it up when she stated, “Language is important, but if you care about impact, it’s really about what it takes to put these words into action, those policy-to-practice connections. It matters who’s championing the work in each country and what supports and partnerships they need to expand adoption and use of OER, and how they create the conditions so that open resources and practices translate into meaningful differences in how students learn and see themselves as learners.”

Unsurprisingly, the theme of the panel was the human center of education. Panelists encouraged us, even as new technologies emerge, to stay true to the principles of Open Education, where a pluralistic and communal approach to learning best supports the encouragement of critical thinking and collaboration in curriculum design by and for communities of learners. At the end of the day, education is about bringing communities together. As Dr. Green aptly said, “Policies don’t stand alone.”

The Legacy & History of Open Science

To learn more about the history of the open movement as it relates to science and open access, we heard from experts Melissa Hagemann, John Wilbanks, and Dr. Vinodh Ilangovan, moderated by CC’s Monica Granados.

Open Science has reshaped how research is created, shared, and accessed. The panel highlighted how open approaches to scientific knowledge can accelerate discovery, increase transparency, and make research more accessible beyond traditional academic systems. From open data to open access publishing, the movement reflects a belief that knowledge created through collective effort should be available to support collective progress. 

One striking example of open sharing leading to collective progress was one Wilbanks shared of the Tres Cantos Antimalarial Set (TCAMS) data set. The dataset contained over 13,000 chemical compounds that had been established as bioactive against the malaria organism. The stewards of the dataset released them under a CC0 license, waiving all copyright and future patent rights. A decade of building on that open dataset led to a number of those compounds now moving toward human testing for the treatment of malaria in a completely open-source drug discovery project. “And it all started upstream with ‘We waive IP,’” Wilbanks said.

The open science movement predates CC, but the critical infrastructure that the CC licenses provided led to something bigger. Dr. Ilangovan shared, “In a bigger sense, normalizing the idea that all research outputs should be made reusable by default, that is what CC has done as a bigger milestone. Because that also opens up a larger area for new scholarship. For example, data analysis on genomes wouldn’t have been possible if there was no CC license on those datasets.”

CC licenses were the legal tools that enabled frictionless sharing. And, as the panelists pointed out, this infrastructure is often one that is taken for granted. CC licenses have done their job so well that they have become nearly invisible. “Just as roads and buildings are maintained,” Monica Granados stated, “we should be maintaining this infrastructure that so many things are built on top of.” John Wilbanks agreed, adding that the future of open science in the next 25 years will depend on that very infrastructure. As large language models pull from the content that has been digitized, “at the end of the day, that infrastructure on which [AI] runs is going to be vital.”

For all of the panelists, the goal of the open science movement has never been to make research open simply for the sake of being open. It has always and will continue to be about the progress and evolution of science, of global knowledge, and of humanity itself.

The Legacy & History of Copyright & Open Licensing

Discussing the history of copyright and & open licensing were Tyng-Ruey Chuang, Kat Walsh, and Tomoaki Watanabe, moderated by CC’s Diyana Noory. 

The legal experts spoke to the early development of the CC licenses and how they changed the relationship between creators and the public by offering a flexible alternative to traditional copyright. Rather than simply asking what people could not do with creative works, open licensing created a framework for creators to define how their work could be shared, reused, and built upon. The panel reflected on how these tools grew from legal infrastructure into a global movement built around collaboration, attribution, and reciprocity.

Some of the biggest challenges of creating the licenses became their biggest strengths: translation, interoperability between legal jurisdictions, balancing simplicity with legal enforceability. Tyng-Ruey Chuang spoke to the challenge of even translating the name Creative Commons into the traditional Chinese they use in Taiwan. The final translation, Chong Yong, translates to “create” and “use,” pointing to the cyclical nature of the licenses. They enable sharing which enables creation which enables further sharing.

Of course, translation involves more than just the literal translation from one language to another. Each license needed to be localized for each legal jurisdiction as well. CC utilized local experts for this process, both a practical decision and one key to the licenses becoming a global movement. The local experts that assisted with the project became local experts in their use—experts who could answer questions specific to their communities’ needs and advocate for their adoption.

And while it is difficult to pinpoint exactly what ingredient in the licenses led to their evolution into the standard for sharing worldwide, that fact cannot be denied. Tyng-Ruey Chuang mentioned his surprise upon learning that CC licenses had become so standardized that they were being taught to the next generation in their Intellectual Property Protection classes. “CC licenses have become the vocabulary to exchange the ideas of sharing.”

The Legacy & History of Open Culture

For our Open Culture panel, we were joined by panelists: Medhavi Gandhi, Merete Sanderhoff, Andrea Wallace, and Giovanna Fortenelle, with CC’s Brigitte Vézina providing opening remarks, and CC’s Dee Harris moderating. 

The conversation touched on numerous topics including how museums, libraries, archives, and cultural institutions have embraced openness to make heritage more accessible and meaningful. Much of that shift was made possible in practice by CC0, CC’s public domain dedication tool, which gave institutions a way to release rights at scale instead of negotiating reuse terms with each individual partner or researcher. Panelists reflected on the shift from institutions seeing collections as assets to control toward recognizing their role as stewards of shared cultural resources. “We have in our DNA the obligation to be for everyone,” Merete Sanderhoff from the National Gallery of Denmark stated. “We are merely the stewards of the collections we hold. They’re not ours to own but we take care of them on behalf of the public.”

Merete continued by sharing that openness began with a simple question: If cultural collections could be shared with one organization, why not share them with everyone? The conversation also explored the responsibilities that come with openness,  including ensuring that communities are respected, living traditions are protected, and new technologies like AI are developed in ways aligned with public values. She concluded: “Opening your collection and your data allows and invites users to interact freely with it and share their knowledge back, thus enriching all of us.”

Giovanna Fontenella shared the story of Madam X, the 19th century painting by John Singer Sargent, that inspired an iconic Hollywood dress and demonstrated how that story had only become visible once that artwork held by The Metropolitan Museum of Art was shared widely on Wikipedia. 

Andrea Wallace made the case that opening up heritage collections was not as scary as it may sound and that there were ways to reconcile open and free access to heritage with an institution’s revenue generating activites. She also emphasized the need to appropriate care of heritage and for ensuring respect for Indigenous rights and protocols.  

And Medhavi Gandhi pointed to the power of open to help communities reclaim their heritage and restore the narrative, calling attention to remaining gaps, including the digital divide and the need to break down linguistic barriers.

Continued Celebrations

Thank you to all of our wonderful panelists, participants, and CC team members who contributed to these rich conversations about CC’s beginnings and the legacy and history of our work over our first 25 years. In the coming months we’ll host two more acts of this series centering CC’s community and the future of Creative Commons and sharing.

We look forward to sharing more events with you soon. Stay up to date on what’s coming at creativecommons.org/events and through our newsletter at mail.creativecommons.org/subscribe. 

The post The Legacy of 25 Years of Creative Commons appeared first on Creative Commons.

Twenty-five years ago, a small group of people made a bet. They believed that if you gave creators a simple set of tools and licenses in language that a lawyer, a machine, and a human could all read, millions of people might choose to share their work with the world instead of locking it down.

“We said, oh gosh, if a million people use these licenses, that would be amazing,” Hal Abelson recalled, laughing at how small that ambition sounds now. Today the number “starts with a B.”

Hal’s quote was one of many memorable moments during Creative Commons’ Founders Fireside Chat, a special benefit event for CC’s 25th anniversary.

Moderator Glenn Otis Brown, CC’s CEO from 2002 to 2005 and the mind behind “Some Rights Reserved,” opened the conversation to three of the people who were there in the beginning: Lawrence (Larry) Lessig, the co-founder and emeritus board member since 2014; Hal Abelson, the founding director who served on the board until 2015 and helped shape CC’s role in open science and OER; and Molly Van Houweling, CC’s founding executive director, who later chaired the board from 2016 to 2022.

What followed wasn’t just a recap of history. It was closer to eavesdropping on old friends remembering how a hopeful guess became the way the world shares.

First, There Was Uncertainty

Molly set the scene first. The early days weren’t defined by confidence, they were defined by not knowing whether the right tool was a license, a search engine, or something else entirely. Not knowing if anyone would actually be part of what was being built. Some early management consultants even produced a report asking why anyone would give away rights they automatically held. Molly’s response still lands: “What’s more motivating than proving the management consultants wrong when they say humans don’t behave this way?”

Then Larry shared an early memory. Creative Commons was incubated out of Stanford Law School, where Larry was a professor running the Center for Internet and Society, and where Glenn, as CC’s first CEO, was building the organization day to day. Larry described coming in one morning and finding Glenn asleep under his desk, having worked through the night out of sheer conviction. That, Larry said, was the moment he realized they’d accidentally lit a spark of entrepreneurial energy that was going to take off everywhere.

Finally, Hal traced the technical DNA of it all back to MIT OpenCourseWare, which launched alongside CC and became what the group nicknamed “the anchor tenant.” This was proof, early on, that serious institutions would use these licenses. “It’s not only putting out to the world,” Hal said. “It’s teaching the world how to make these things.”

A Voice from the Past

Partway through the conversation, a familiar name lit up in the Zoom window. Eric Saltzman, an original CC board member who’d been part of the project before it was even incorporated, unmuted with a cheerful “Hi, everyone.” He brought a layer of stories that was equal parts mischief and delight, like the time IP attorney Diane Cabell quietly cleared CC’s now-iconic logo of any conflict with Chanel’s, simply by pointing out that one mark’s two C’s curved one way and the others curved the opposite. Then he shared a story fresh off a CC summit in Korea where he found himself deep in the countryside, swapping stories with a complete stranger from Austria who had no connection to CC at all. On hearing that he had just come from “a Creative Commons thing,” the stranger lit up, describing it as one of the best things that had ever happened in his work. Half a world from where any of it began, a stranger had been impacted by CC tools and licenses.

A Candlelit Apartment and a Supreme Court Loss

Some of the richest moments were the ones that simply hadn’t been told in a while. Larry resurrected the story of meeting Gilberto Gil, Brazil’s eventual Minister of Culture, when his friend John Perry Barlow brought him along, at one in the morning, to Gil’s apartment. They found Gil sitting on the floor by candlelight, guitar in hand, seemingly paying them no attention at all. That didn’t stop Larry from pitching the idea behind Creative Commons anyway, right there as Gil kept playing. Then, suddenly, Gil stopped. He looked up, and said, “I love this, this is great.” That encounter helped launch CC Brazil and led, years later, to Gil performing at CC’s fifth anniversary celebration.

And near the end, Larry shared something he said most people don’t know. The real origin of his drive to start CC traces back to a promise made to Eric Eldred, the plaintiff in a Supreme Court case Larry argued, and lost seven to two, that challenged Congress’s power to keep extending copyright terms. Before that case began, Larry told Eldred that no matter what happened in court, they’d build a commons for creative work regardless. Creative Commons, in Larry’s words, was “the delivery on that promise.”

AI Is In the Room

No CC founders’ conversation in 2026 could avoid the question in every room these days. Glenn put it to the panel directly: “Today, knowledge and culture is being reused on an almost unimaginable scale by machines, by models, by LLMs, some in a handful of private hands, some more distributed in open ways. How do you begin to process the notion of reusing creative work, building on what came before, in a 2002 notion versus today? Do you think of Creative Commons culture, or remix culture, the same way in a world of AI, or do you think about it differently?” 

Larry argued for drawing a clear line between content built for humans to use and content built for machines, and pushed back hard on the idea of extending legal protections to AI systems like we do to people. 

Molly offered the conversation’s emotional center of gravity, stating that remix culture and CC’s success stories have always mattered because they connect humans to other humans, and that fully machine-made work risks crowding out, or making us doubt, that sense of connection altogether. 

Hal turned to the lawyers with a question. “Now we’re talking about the sound of somebody’s voice, we’re talking about the style by which somebody does something and we’re actually seeing lawsuits that are claiming copyright infringement over that. I was wondering how you think about that expanding as the machines can cover more and more stuff.”

The answer, as Glenn pointed out, is unsettling. Style alone has never been something copyright protects. Working “in the style of” someone else has always fallen on the unprotected, idea side of the law. The trouble is that mimicking style is exactly what AI does best, which is why the conversation keeps drifting toward whether other legal tools might end up doing more of that protective work than copyright ever could.

Closing: One Word Each

Asked to sum up 25 years in a single word, the answers landed like a thesis statement for the event. Hal went with empowering. “The reason anyone’s doing this is that we’re empowering other people to build on existing work.” Eric, laughing that Hal had already used most of the good words, added inspiring. Larry kept it simple: community. Molly chose serendipity, “because when you release your works into the world, there’s no telling who, with whom, they will connect.” And Glenn closed it out with belonging,  “a thing that you belong to, whether you actually realize it or not.”

Twenty-five years later, no one described a master plan. What came through instead was a string of unplanned connections, a commitment to the values of openness, and a determination to find a better way to share. A man asleep under a desk. A guitar that stopped mid-song in a candlelit apartment. A stranger in the countryside. A promise kept after losing a case at the Supreme Court. Somehow, that’s how you get from “a million people, that would be amazing” to a number that starts with a B.

If you missed the conversation, you can watch the recording. We have more to come as part of our 25th anniversary. Stay up to date about upcoming events and subscribe to our newsletter so you don’t miss a thing.

The post Inside the CC Founders Fireside Chat appeared first on Creative Commons.