In an era where Artificial Intelligence (AI) is no longer a science fiction scenario but the primary engine of the global economy, the recent intervention by the French newspaper Le Monde poses the most critical question of our decade: Who holds the keys to ethics? The observation that we cannot allow private companies to act simultaneously as "judge and party" highlights a profound crisis of institutional trust. As Silicon Valley and Shenzhen giants accelerate the development of models approaching Artificial General Intelligence (AGI), political power appears to be panting behind, trying to regulate something it barely understands.
The Paradox of Self-Regulation
The history of technology is riddled with examples where industry was invited to self-regulate, usually with disastrous results for the public interest. In the case of AI, the problem is intensified by the complexity of the algorithms. When a company like OpenAI or Google develops a new model, it is the one conducting the safety tests ("red teaming"), it defines the boundaries of the system's "ethical" behavior, and it decides what to reveal to the public. This lack of independent oversight sets a dangerous precedent. "Safety" often translates into "corporate compliance," and "ethics" into a mechanism for protection against legal liabilities, rather than the defense of human rights.
Furthermore, the concentration of power in a few hands means that the values embedded in these systems—from political neutrality to cultural sensitivity—reflect the worldviews of a specific elite. Le Monde rightly points out that shaping the future of humanity cannot be the side effect of maximizing shareholder profit. We need a democratic legitimacy that only state and supranational institutions can offer, provided they possess the necessary technical expertise.
The Illusion of "Safety" as a Barrier to Competition
One of the darkest aspects of the current rhetoric regarding "AI risks" is how it is used by dominant players to raise barriers to entry. By arguing that AI is so dangerous that it requires draconian and expensive regulatory frameworks, large corporations are essentially asking the state to ban competition from smaller startups or the open-source community. This is the classic phenomenon of "regulatory capture," where regulations end up serving those they were supposed to control.
- Promoting closed models in the name of safety limits transparency and scrutiny by the academic community.
- State subsidies for developing "safe" AI often end up with the same companies that created the problem.
- The lack of standards for training data use violates the copyrights of millions of creators without compensation.
Towards a New Social Contract for the Digital Age
To address this power imbalance, a radical paradigm shift is required. AI regulation must not be limited to technical specifications but must extend to the political economy of technology. This means establishing independent public bodies with the power to conduct audits on source code and training data, without being bound by trade secret clauses when the public interest is at stake.
"Technology is too serious a matter to be left exclusively to technocrats and investors," the analysis notes.
The European Union, with the AI Act, took the first step, but its implementation remains a battlefield. Lobbying pressures are stifling. The challenge for 2026 and beyond is the creation of an international framework—a sort of "IAEA for AI"—that will ensure these systems are developed with human well-being in mind rather than algorithmic dominance. If we fail to set the rules now, we risk living in a world where justice is dispensed by private software, and truth is defined by who owns the largest servers.
