Anthropic and AI: A Technological Advancement Reserved for an Elite

⚡

Key Takeaways

1Artificial intelligences are built through the interactions of millions of users, but their benefits are concentrated in the hands of a few companies.

2Anthropic, valued at $965 billion, has launched a powerful AI model, but it is reserved for privileged partners, raising questions about equitable access to technology.

3The digital divide is widening, leaving public infrastructures and less developed countries vulnerable to cyber threats.

💡Why it matters — Unequal access to AI could exacerbate global inequalities in security and technological development.

AI: A Collective Project for Limited Benefit

Modern artificial intelligences, although they appear to be autonomous and sophisticated tools, are actually the result of immense collective effort. Every interaction we have with these systems contributes to their improvement. Users, often unknowingly, help refine these technologies by sharing their preferences and correcting errors. However, this contribution, while essential, is rarely recognized or compensated. Companies developing these AIs highlight the option to disable this sharing, but it remains a choice that few people activate.

In the professional world, the use of these tools has become almost indispensable. Refusing to adopt them can be perceived as a lack of competence, which can have consequences for an individual's career. Thus, although the use of AI is technically voluntary, it is increasingly imposed by the expectations of the job market.

A Technology Owned by a Handful of Players

The apparent effectiveness of AI models does not guarantee their reliability. These systems can produce information that seems credible but is not always verified. True reliability depends on the data and systems established around these models to verify and trace information. However, these verification systems are often commercialized, creating a barrier to access.

This situation creates a divide between those who have access to verified data and those who do not. For example, an average citizen will only have access to information available on the public web, while a professional like a lawyer or a doctor can access specialized databases. Institutions that can afford these services will have access to more reliable systems, leaving others in uncertainty.

The Mysteries of AI Models

Artificial intelligence models, much like ancient pyramids, possess internal aspects that remain misunderstood, even by their creators. The interpretability of these systems is a major challenge, as developers themselves do not always understand how decisions are made within these models.

In June 2026, several significant events highlighted this complexity. On June 1, Anthropic filed for an IPO after being valued at $965 billion. A few days later, on June 4, the company called for a coordinated pause in AI development, before launching a powerful model on June 9, reserved for verified partners. This series of actions demonstrates how companies can control both the acceleration and regulation of AI, raising questions about the fairness of this access.

The Consequences of Restricted Access

The AI model launched by Anthropic, capable of identifying vulnerabilities in major systems, is reserved for an elite. This restriction of access means that the best cybersecurity capabilities are held by a private company in collaboration with a state, leaving many public institutions and developing countries without adequate protection.

The consequences of this inequality are concerning. Institutions that cannot afford these technologies remain vulnerable, while those that do have access gain a significant advantage in security. The central question is no longer just who can use these models to attack, but also who can defend themselves effectively.

Towards a Public Interest AI

To address these challenges, it is crucial to treat AI tools as fallible assistants, requiring verification and cross-checking. However, an individual response will not be enough to bridge the current gaps.

The solution lies in a collective approach. Public infrastructures, such as libraries and universities, should include AI systems accessible to all. The goal is not to compete with private innovation but to ensure that access to reliable information and protective capabilities does not solely depend on wealth or relationships with private companies.

Initiatives already exist, supported by UNESCO and the United Nations, to develop public interest AIs. Projects like OpenEuroLLM are working to create accessible models for European languages, but it is crucial to extend these efforts to other languages and cultures.

The pyramids of AI, like their ancient counterparts, may well stand for a long time. However, to prevent the collective work that built them from being forgotten, it is essential to build a system that benefits everyone, not just a privileged minority.