Will we soon need to show our passport to use AI?

TL;DR

Artificial intelligence is generally analysed in terms of model performance, the vast investments required to develop it or the competition between laboratories. Yet another transformation may be under way, discreet but potentially even more consequential: control over access to the models themselves.

The agreement recently reached between Anthropic and the Trump administration sets a precedent in this respect. After two weeks of tensions, during which Europe and Silicon Valley discovered a new sovereignty problem, the US Department of Commerce authorised the gradual restoration of access to the Mythos 5 model for some of the company’s customers. In return, Anthropic agreed to strengthen its model safeguards and work with Washington on protocols governing future generations of artificial intelligence.

The episode could be viewed as a simple compromise. In reality, it reveals a much more significant development: the governance of frontier models is no longer solely a matter for laboratories. It is gradually becoming an issue of industrial policy, national security and, de facto, technological sovereignty.

The Anthropic case marks the end of an era

Under the new agreement, Mythos 5 could gradually be restored for a limited group of authorised customers, while negotiations concerning Fable 5 continue.

In a letter to Anthropic, Commerce Secretary Howard Lutnick reportedly indicated that the company had agreed to work with the US government on protocols applicable to future models.

Until now, the release of a new model was handled almost exclusively by the laboratories’ internal teams: research, safety, red teaming, alignment and then commercialisation. The US administration now intends to participate earlier in that process.

The precedent also extends far beyond Anthropic. OpenAI has announced that GPT-5.6 Sol will initially be restricted to a limited number of organisations approved by the US administration. Meta has been encouraged to submit its models voluntarily for federal evaluation. Google, Microsoft, xAI and Anthropic are already participating in the work of the Center for AI Standards and Innovation, which is responsible for evaluating the most advanced models.

For several months, laboratory executives, cloud providers, cybersecurity specialists and US government representatives have discussed, often informally, the need for stricter controls on access to frontier models. These discussions, until now largely invisible, are becoming public. The Anthropic case provides the first concrete example, centred on one question: who should be allowed to access the most advanced models?

Towards a KYC system for artificial intelligence

The idea is hardly revolutionary and already exists in other industries. Banks apply Know Your Customer procedures to identify their clients. Cloud providers rely on sophisticated identity and access management systems. The defence sector has long operated through different levels of security clearance.

Artificial intelligence could adopt similar mechanisms. Access to the most powerful models would no longer be universal. Depending on the level of risk associated with a model, users might be required to verify their identity, organisation, country of origin, business activity and, in some cases, the intended purpose of their project.

A model specialising in offensive cybersecurity, computational biology or autonomous agents could require different levels of authorisation depending on the applicable regulations.

The principle would not be to prohibit models, but to establish several tiers of access comparable to the security-clearance levels already used by government agencies and critical infrastructure operators.

From a practical standpoint, this development would require no technological breakthrough. The necessary infrastructure already exists. Microsoft operates Azure Government, Azure Secret and Microsoft Entra to manage identities and access in sensitive environments. Amazon offers GovCloud, Secret Region and a complete range of IAM services. Google provides Cloud Identity and Assured Workloads for regulated environments.

In other words, the required technical components are already operational. They would simply need to be applied directly to artificial intelligence models.

The distribution cycle for a model would gradually shift from a “release first” approach to a “clearance first” model, requiring laboratories and other AI companies to adapt.

A new control layer is emerging

This development would profoundly alter the industry’s balance of power. While laboratories would retain control of their models, hyperscalers could become the real gatekeepers of their distribution.

Microsoft, Amazon and Google already control user identification, clearance management, audit logs, sovereign environments and infrastructure certified for government and defence use. They could therefore be tasked with enforcing these controls.

This approach is gradually moving frontier models towards the regulatory regime applied to dual-use technologies. They are ceasing to be ordinary software products and becoming critical infrastructure, comparable in some respects to certain space, nuclear or military capabilities.

Large corporations will probably need to introduce a new governance framework for model access. IT departments would be required to manage specific authorisations, log usage, audit interactions with certain models and demonstrate compliance during regulatory inspections.

A new market could therefore emerge around “AI Identity Management” and “AI Access Management”, at the intersection of cybersecurity, regulatory compliance and cloud computing.

This development also illustrates the pragmatic American approach to AI governance. Rather than concentrating regulatory efforts on model transparency or documentary obligations imposed on developers, Washington appears to favour a more operational lever: controlling the distribution of, and conditions of access to, the most advanced models.

Future models may be designed to satisfy regulators

One consequence remains largely absent from the debate. If market access eventually depends on government approval, laboratories will need to incorporate that constraint into the design of their models from the outset.

Safety and alignment teams will no longer work solely to reduce hallucinations, limit malicious uses or improve system robustness. They will also need to meet regulatory requirements that could determine whether a model is authorised for commercial release.

Future models could therefore be designed not only to maximise performance, but also to satisfy compliance criteria defined by public authorities. Frontier models may no longer be merely “safety by design”; they may also need to become “regulation by design”.

This change could prove even more consequential than the next technological breakthroughs. For the past two years, global competition has been viewed as a race for parameters, GPUs and gigawatts. The Anthropic case suggests that another contest is now beginning: the battle to control access to the most advanced AI models.