Hundreds of millions of people train and improve AI models with every conversation. Without knowing it. Without compensation. Without even being informed. Projet Pollen was born to change that.
A Story of Flowers and Bees
To understand what Pollen means, we have to begin by setting aside the usual metaphors. AI is not the tool we use. Nor is it the service we consume. Those two words are inadequate to describe what truly happens when a human being enters into dialogue with a machine.
We should think instead of one of the most beautiful love stories nature has ever invented: the one between flowers and their pollinators.
The bee doesn’t know it’s making flowers evolve. It’s looking for nectar. That’s all. The flower doesn’t know it’s shaping the bee. It attracts what will fertilize it. That’s all. And yet, over millions of years, they have sculpted one another. Flowers developed colors, shapes, and scents calibrated for their visitors. Bees refined their vision, their anatomy, and their behaviors to make use of those resources. Every visit carries pollen — and with it, genetic information that transforms the entire species.
This is exactly what happens between users and artificial intelligences. We are not mere consumers of a service. We are the pollinators who allow AI to adapt, correct itself, and co-evolve with the real world.
Every reformulated prompt, every correction made, every piece of feedback given is a grain of pollen. Deposited without a thought. Harvested without count.
Every reformulated prompt, every correction made, every piece of feedback given is a grain of pollen. Deposited without a thought. Harvested without count.
The Lopsided Contract
Here is where things become complicated. This massive contribution remains largely invisible. And above all, insufficiently recognized.
When you interact with generative AI, you are both customer and supplier of raw material. You pay — or you accept targeted advertising, which amounts to the same thing — to access a service. And in the same breath, you feed an improvement loop that benefits the company billing you.
It’s a little like paying admission to a botanical garden where, besides taking a walk, you help pollinate rare species simply by being there. With no mention on the ticket. With no share in the sale of the seeds.
This recognition exists, of course. But it remains deeply asymmetrical. Anonymous — your contribution dissolves into an ocean of aggregated data. Unpaid — the models trained on these interactions are worth billions, while you receive the service you paid for, no more and no less. Uncontracted — when you accept the terms of use, you waive certain rights with no room for negotiation.
We have already lived through this scenario with social networks. Facebook, Instagram, and YouTube built empires on content generated for free by their users. Two decades later, the record is mixed: a few creators have prospered; the majority have enriched the platforms without proportionate return.
Generative AI reproduces this pattern with a new intensity. Here, what is used is no longer only content — it is a way of thinking, phrasing, and reacting. Something more intimate. The pollen you deposit is a fragment of your intelligence.
The pollen you deposit is a fragment of your intelligence.
Five Principles to Set It Right
Projet Pollen is not a petition. Nor is it an indictment. It is a structured, well-reasoned, legally grounded proposal — a framework of principles that AI companies could adopt voluntarily, and that regulators could later strengthen.
Five principles form its backbone.
Transparency. Users have the right to know what happens to their conversations. Are they used to train the next models? How long are they kept? With whom are they shared? The answers exist in the terms of use, often buried in pages of jargon. They should be readable. Accessible. Up to date.
Recognition. The contribution of users must be named for what it is: an input that creates value. Not merely anonymous data aggregated into a flow. To recognize is not to compensate — it is first to give a reality its proper name.
Informed Choice. Agreeing to contribute should be a voluntary act, not a clause buried inside a pre-checked box. Default consent is the opposite of informed consent. Every user should be able to activate or deactivate their participation in model training, and that choice should be able to change over time.
Reciprocity. When a contribution creates economic value, some form of return is legitimate. That return can take several forms — privileged access to new model capabilities, profit-sharing, better-protected personal data, or simply an improved service for active pollinators. Pollen does not prescribe a single form. It asks that some form exist.
Shared Governance. The choices that shape AI models involve millions of people. Those people should have a voice in the matter — not individually, but collectively, through representative bodies. User advisory councils already exist on some platforms. They should become the norm.
These five principles do not come from nowhere. They are anchored in the GDPR, which has laid the foundations of European personal data law since 2018. They also speak to the European AI Act, adopted in 2024 and now governing artificial intelligence systems across the continent. They extend these frameworks rather than contradicting them.
To recognize isn’t to compensate — it’s first to put the right words on what is real.
The Unexpected Convergence
During the drafting of the Pollen dossier, a surprise emerged.
In January 2026, Anthropic published Claude’s Constitution — a founding document setting out the values and principles guiding the development of its artificial intelligence. When we read it, we recognized our own project. The same conclusions, reached from opposite directions — they from the side of the machine, we from the side of the user.
Transparency about how models work. Honesty in interactions. Respect for human autonomy. Explicit recognition that users are not mere passive consumers. All these principles appear in Claude’s Constitution — and they answer Pollen’s concerns point by point.
This convergence is no accident. It suggests that the question is not whether we should recognize users’ contribution to AI, but when and how. AI pioneers like Anthropic have already understood the issue. Others will follow, because they will have no choice. What remains is to accelerate the movement — and to make sure that the proposed principles truly protect the pollinators, not merely the image of the companies.
The Dossier That Flew to San Francisco
In March 2026, a complete dossier was sent to Anthropic’s founders. Manifesto. Legal note. Draft charter. Ten days of work condensed into a package that flew from a small rural town in Nouvelle-Aquitaine, in southwestern France, to San Francisco.
Signed by an unknown author. Without an audience. Without institutional backing. Without any other legitimacy than that of a clear-eyed conviction: the choices we do not make today will set the precedent for decades.
We do not know what will become of this dossier. Perhaps nothing. Perhaps a polite acknowledgment. Perhaps a real conversation. What matters is not the destination — it is having asked the question clearly, with legal rigor, with the quiet strength that the law provides.
Pollen is not a project waiting for an answer. It is a project building, brick by brick, a shareable framework. Because if Anthropic does not take it up, perhaps another actor will. Or a regulator. Or a citizens’ collective. It does not matter who holds the final pen — what matters is that the principles exist, documented, available, and ready to be taken up.
The choices we don’t make today will set the precedent for decades to come.
What You Can Do
Know that you are a pollinator. That is already a great deal. Become aware that with every prompt, every correction, every conversation, you deposit a grain into the collective memory of machines. This is not a burden — it is a shared responsibility, finally named.
Demand transparency. Ask your AI providers what they do with your conversations. Read the terms of use. Activate the options that allow you to check how your data is used. Refuse those that do not suit you.
Talk about the subject. Questions of AI governance are too often left to lawyers, regulators, and product teams. They concern us all. The more they are discussed outside specialist circles, the more collective pressure for fair practices will become irresistible.
Follow the work in progress. The project’s publications, articles, and updates appear as the work unfolds. Pollen is an open project, and its developments remain accessible to those who want to follow them.
You are a pollinator. Perhaps it’s time you knew. And acted accordingly.
The articles that extend Pollen — reflections, news, and legal convergences — appear on our blog: https://neurostratum.com/index.php/blog/
