Manifesto for the Responsible Use of AI in Drupal Projects

In a context where AI is often adopted opportunistically, driven by trends or competitive pressure, we choose a critical and responsible approach. We reject blind use of AI, disconnected from real-world use, its environmental, social, and democratic impacts, or the concrete value it brings to end users.

Our thinking is based on a clear analysis of the transformations that AI imposes on the web. By settling at the heart of digital experiences, it raises fundamental questions: diversity of information, uniformity of content, dependence on platforms, concentration of technological power, and the worsening of the digital world’s ecological footprint. We are convinced that AI must not weaken the web, but on the contrary, strengthen it by enriching uses without impoverishing the plurality of voices or undermining the economic and democratic models that support it.

This manifesto sets out the principles that structure our approach. It is built around four commitments:

• truly open source AI, natively integrated into Drupal and based on technologies we control;
• AI used responsibly in light of environmental issues;
• sovereign AI, limiting technological and legal dependencies;
• French AI, embedded in a European ecosystem consistent with our values.

Open source is at the heart of our DNA. As a historic Drupal agency, we built our expertise on a culture of openness, sharing, and contribution. This model has shaped the web as we know it: a space for collective innovation, transparency, and digital freedom. We are convinced that artificial intelligence should not stray from these principles.

Our approach to AI is based on two inseparable pillars:

• Drupal, as the open source foundation structuring usage, content, and workflows;
• the choice of the AI model, which is independently selected, interchangeable, and fully controlled.

AI natively integrated with Drupal

At bluedrop.fr, artificial intelligence is not an afterthought or an external tool bolted on to projects. It is integrated at the very core of Drupal, the open source CMS underpinning the sites and platforms we design.

This integration is based on a Drupal recipe, a key concept in the Drupal ecosystem. A recipe is a coherent combination of modules, configurations, views, blocks, and best practices designed to meet a specific need. Unlike a standalone module, it offers a comprehensive, well-documented, and ready-to-use vision, aligned with the standards and philosophy of the CMS.

This approach makes it possible to:

• build upon what already exists rather than multiplying custom developments;
• ensure the maintainability and scalability of projects;
• embed AI within a clear, understandable, and manageable technical framework.

Drupal AI: Contextualized AI, not just an external tool

The Drupal AI recipe enables the direct integration of artificial intelligence capabilities into the Drupal admin interface, where editorial and technical teams already work.

AI thus becomes a CMS feature, contextualized by content, user roles, and the specific technical requirements of Drupal projects. It supports teams with targeted uses: advanced search engines, conversational assistants, content generation, workflow automation, and even SEO optimization and Generative Engine Optimization (GEO).

For developers, it is also an enabler for simplifying low-value-added tasks: assistance during complex content imports, help with the configuration of editorial workflows, generation or adaptation of configurations, as well as support for certain maintenance and update operations.

These actions are never performed “in a vacuum”: they take into account content structure, editorial purpose, Drupal architecture choices, and the publication environment. AI suggests, assists, and guides. It does not replace human decisions or the technical expertise of teams.

The choice of the Mistral AI Small model: openness and control

Here, Drupal plays a decisive role: it does not lock you into a specific AI model. The CMS allows different models to be interfaced, whether open source or hosted on managed infrastructures. This replaceability is a fundamental lever for sovereignty and accountability.

It is in this context that we chose the Mistral AI Small model, not by default, but because it coherently meets our criteria: openness, auditability, efficiency, and technical control. This choice remains reversible, scalable, and governed by the real needs of projects—not by dependency on a single provider.

By combining Drupal as an open source framework and the AI model as an interchangeable component, we avoid black box approaches, technological lock-in, and structural dependence. This combination forms the basis for our vision of truly open source AI.

Artificial intelligence is not a neutral technology. Its development relies on heavy, energy-intensive infrastructures that consume large amounts of resources. Training large language models mobilizes thousands of GPUs over several weeks and generates CO₂ emissions that can reach several hundred tons for a single training cycle. Added to this is the impact of inference—that is, the daily use of these models—which is now becoming a major source of energy consumption as adoption becomes widespread.

These observations raise a fundamental issue: AI, as it is being massively deployed, comes into direct conflict with the goals of digital sobriety and ecological transition. As actors in the web ecosystem, we have a duty to acknowledge this reality without sugarcoating it. Yes, AI contributes to the increasing environmental footprint of digital technology. And no, these issues are not yet fully resolved.

At the same time, we face a structural dissonance. Usage patterns are evolving rapidly, and the expectations of organizations and users are turning to AI; to refuse to respond would mean disconnecting ourselves from market realities and losing the ability to effectively support our clients. This tension between ecological responsibility and the need to remain competitive is real, and we choose to acknowledge it rather than deny it.

Our response is neither the rejection of AI nor its indiscriminate adoption. It is based on a simple principle: if AI is used, it must be used with the highest level of caution possible. This means limiting use to cases where the added value is proven, rejecting gimmicky deployments, and above all, choosing technically more efficient solutions.

This is the logic behind our choice of the Mistral AI Small model. Unlike massive, general-purpose models, this model is designed for targeted uses and offers a much better performance-to-energy-consumption ratio. Its smaller size allows for a significant reduction in inference costs, sometimes by an order of magnitude compared to larger models, while remaining perfectly suited to editorial and business use cases we address in Drupal.

This choice is also consistent with our overall approach to eco-design. We design efficient websites, limit technical complexity, optimize performance, and favor low environmental impact hosting, notably via infrastructures that incorporate circular economy principles for cooling. AI is no exception to these principles—it is integrated within a framework already constrained by strong environmental requirements.

Ultimately, our stance is deliberately exacting. AI is neither a miracle solution nor a trivial technology. It is a powerful tool with real impacts, which we choose to use with restraint, lucidity, and responsibility, striving to reduce its footprint as much as possible, while never losing sight of the ecological limits that digital technology now faces.

Digital sovereignty is now a major democratic and societal issue, especially in the field of artificial intelligence. AI is no longer just a technical tool: it shapes content production, access to information, automated decisions, and ultimately, part of the public debate. In this context, the choice of technologies and players to whom we entrust these functions is far from neutral.

The recent history of digital technology has already shown us the consequences of an excessive concentration of technological power. By allowing American players to establish a virtual monopoly on social networks, we have seen major abuses emerge: algorithmic opacity, the amplification of fake news, the spread of conspiracy theories, and interference in democratic processes. These platforms, which have in effect become information infrastructures, now influence public opinion without real democratic oversight.

This concentration also carries operational consequences. In recent years, many examples have illustrated the possible pitfalls of a digital quasi-monopoly: unilateral changes to API terms of use, sudden increases in access costs to services that have become essential, feature restrictions without consultation, or even the outright discontinuation of certain products. These decisions, made by dominant players, are imposed on businesses and institutions that depend on them, without any real checks or balances.

In the AI field, these mechanisms are even more sensitive. When an organization relies exclusively on a proprietary model hosted by a dominant player, it is effectively accepting that its content, data, and sometimes business logic will pass through a platform whose rules and development it neither controls nor understands. Recent changes in pricing policies or operational conditions for some AI APIs make this risk very clear: services that were previously accessible suddenly become costly or limited, putting users in a position of strong technological dependency.

On top of these issues are additional legal and democratic dimensions. Extra-European legislation, such as the American Cloud Act, can theoretically allow access to data hosted by companies subject to these legal frameworks, even when servers are located outside the United States. For public institutions, media, associations, or organizations handling sensitive data, this uncertainty represents a major risk.

Failing to learn from these mistakes would mean repeating, with AI, the same patterns of dependency and loss of control we have seen with social networks or certain proprietary software in public services. In this context, digital sovereignty goes beyond the location of servers: it means having the ability to understand, audit, host, and govern the technologies we use.

This is why we have chosen a model like Mistral AI Small, which can be deployed on French or European infrastructures and is compatible with principles of openness and technical control. This choice aims to preserve the strategic independence of our clients and to contribute, at our level, to a more resilient, pluralistic and democratic digital world.

Finally, the French dimension of our approach is not a minor detail. It strengthens our ability to contribute to a strong and independent European technological ecosystem. By choosing a model developed by a French team, we are investing in local innovation that shares our understanding of European cultural and social values.

This choice not only supports a strategic vision for AI, but also contributes to building a sovereign sector in the face of global competitive pressures. At a time when regulatory frameworks like the European AI Act aim to govern technological developments in a way that is both protective and innovative, choosing a French player aligns with a coherent and forward-thinking approach.

Artificial intelligence is neither neutral nor inevitable. It is the result of technical, economic, and political choices that shape the web, information, practices, and, more broadly, our society in a lasting way. The mistakes of the past (platform concentration, technological dependency, algorithmic opacity, underestimated environmental impacts) now compel us to take greater responsibility.

At bluedrop.fr, we integrate AI because our clients and their needs demand it, but we refuse to adopt it indiscriminately. Every use is questioned, every solution is evaluated, every impact is taken into account. We embrace the tension between innovation and responsibility, and turn it into a requirement.

It is with this mindset that we chose a sober, sovereign, French open-source AI, natively integrated with Drupal and hosted on controlled infrastructures. This choice is neither opportunistic nor ideological: it is the result of a clear understanding of the environmental, democratic, and societal stakes related to AI, and the resolve to remain consistent with our longstanding commitments to eco-design and open source.

We are convinced that the future of the web will not be built on closed, energy-hungry, and uncontrollable technologies, but on tools that are understandable, governable, and aligned with the public interest. In this sense, our approach to AI is not an end in itself, but a stance of active vigilance, serving digital projects that are useful, responsible, and respectful of both citizens and the planet.

This manifesto is an invitation to question assumptions, to reject the easy path, and to work together to build the use of artificial intelligence that truly serves progress, without abandoning our values.