GPT-5.6 and the End of AI’s Belle Époque: The New Battle for AI Control
How GPT-5.6 exposes the divide between U.S. AI governance and Europe’s AI regulation — and why the real battle is now over control, safety, and power.
Table of Contents:
▫️U.S. AI Regulation: Power Over Law
▫️Why AI Safety Now Matters
▫️OpenAI’s Internal AI Governance
▫️Who Should Control AI?
▫️The New U.S. Control Model
▫️AI Competition: Speed vs Safety
▫️From Better Models to Better Governance
▫️AI Compliance for Business
▫️Europe’s AI Policy Through Law
▫️Conclusion: The Battle for AI Control
The real question is no longer which AI model is stronger. The real question is who controls it.
That is the divide now emerging between the American and European approaches to AI regulation — and GPT-5.6 makes it impossible to ignore.
On 8 July, OpenAI unveiled GPT-5.6. But this was not just another model launch. It was a signal that the AI debate has entered a new phase.
For years, every major AI release was treated as a technical race. People asked whether the new model was more powerful, faster, cheaper, more convenient, better at coding, better at reasoning, or better at passing benchmarks.
That era is ending.
Now the central questions are different. Who controls the model? Who decides when it can enter the market? Under what conditions does it become available to users? Who defines acceptable risk? And who regulates its public use?
GPT-5.6 is not only a product release. It is a case study in the new politics of AI governance, AI safety, and AI control.
🟩 The American Approach: AI Regulation Through Power, Not Law
The U.S. government was actively involved in the release of GPT-5.6. OpenAI eventually brought the model to market, but before doing so, the company spent a long time working through the issue with the federal government in a hands-on, ad hoc process.
This matters.
U.S. law does not require a special licence or formal approval to release this kind of AI model. Legally, OpenAI did not need to obtain “permission” from the government. But when money, power, and AI national security are involved, formal legal requirements are rarely the whole story.
This episode shows that the United States already sees powerful AI models not merely as commercial products, but as technologies capable of affecting national security.
And frankly, that concern is justified.
This is where the contrast with Europe becomes clear.
The United States has effectively chosen manual AI regulation: closed consultations, negotiations, testing, and direct interaction between government officials and the company.
The European Union has taken a different path. Europe does not currently have frontier AI companies with the same global scale, infrastructure, and market power as the leading U.S. players. In that sense, Europe is more a consumer of foreign AI products than the main producer of them.
That is why the EU has chosen legal AI regulation: the AI Act, formal obligations, documentation, transparency, risk assessment, AI compliance, and oversight.
Put simply, the United States regulates AI through political and administrative power. Europe regulates AI through law.
🟩 Coding, Biology, and Cybersecurity: Why AI Safety Is No Longer Optional
GPT-5.6 is not just a text model for writing articles, emails, or short answers. Models of this kind have moved far beyond text.
They are about programming, agentic tasks, cybersecurity, vulnerability discovery, automated analysis, and potentially even the creation of prohibited substances or biological weapons.
This is where the dual-use problem becomes unavoidable.
The same model can be useful and dangerous at the same time. It can help security specialists write patches, detect vulnerabilities, and protect systems. But similar capabilities can also be used in the opposite direction: to identify weaknesses, automate attacks, or lower the barrier to dangerous knowledge.
Biology is another example. AI can support medicine, research, diagnostics, and drug development. At the same time, these technologies can be used for dangerous purposes if access is not controlled.
That is why the old argument that “it is just a tool” is becoming less convincing.
The more powerful the model becomes, the less it looks like ordinary software and the more it resembles a strategic resource. I would not call AI a weapon. It is a new kind of resource. Its power is not limited by how much of it exists on the planet, but by how effectively and safely society can use it.
In the right hands, such a resource can be medicine. In the wrong hands, it can become a problem.
This is why AI safety can no longer be treated as a decorative layer added after launch. It is becoming a condition for access, legitimacy, and control.
🟩 OpenAI Is Building AI Model Governance From the Inside
OpenAI is trying to manage risk inside the company itself. It does this through model restrictions, safety mechanisms, request monitoring, access levels, and internal security procedures.
In other words, OpenAI is not simply building a model. It is building an entire control infrastructure around the model.
This is already a form of AI model governance. The company decides how the model behaves, who can access certain capabilities, which requests are blocked, which risks are monitored, and which users or use cases receive more trust.
But that raises the main question: who decides that the model is safe enough?
In the United States, this is effectively decided through interaction between OpenAI and the government. A significant part of that process takes place behind closed doors: without broad public access, without independent journalists, without civil society, and without external experts having real influence.
Society receives the finished product, but has almost no influence over how that product was tested, which risks were considered acceptable, and why the model was allowed into public use in the first place.
Europe has chosen a different path: protection through legal regulation. But Europe faces another problem. It is extremely difficult to regulate a system you do not fully understand.
AI moves faster than legislation. That means there is a real risk that the law becomes outdated before it is even fully applied.
🟩 Who Should Control AI?
At the moment, many of the most important decisions are made by the company itself.
OpenAI decides which risks are acceptable, which requests should be blocked, who receives access, which features should be restricted, which scenarios are dangerous, and which are permissible.
Yes, the company has expertise. But the company also has a commercial interest. Its decisions can be influenced by investors, corporate clients, politicians, government agencies, competitors, and the market.
From society’s point of view, this creates a serious imbalance.
Society did not choose for AI of this scale to become part of everyday life. But now it is forced to live in a world where these models already influence business, education, media, security, politics, and access to information.
This is the missing layer in current AI governance: real AI oversight that is independent, technically competent, and not captured by either governments or corporations.
In my view, the only logical long-term solution is the creation of an independent supervisory body. Such a body must be technically competent enough to understand the technology and independent enough to protect society from abuse by both governments and commercial companies.
Perhaps such a body will be created one day. Until then, we risk remaining puppets in the hands of Big Brother.
For now, the situation looks very different. Governments and corporations have obtained a new strategic resource, but they do not yet fully understand its long-term consequences.
It resembles the moment when the Spanish brought enormous amounts of gold from the New World to Europe, without understanding that this resource could transform the economy and lead to serious consequences.
Something similar may happen with AI. The technology looks like a source of power, profit, and influence. But if it is integrated into society incorrectly, it can create systemic risks.
🟩 The New American Control Model
The American approach does not require a mandatory licence to release an AI model. Instead, a voluntary system of interaction between the government and developers is emerging.
This may include preliminary government access to certain frontier models, testing, risk assessment, and national security consultations.
This is not strict regulation in the classical sense. But it is no longer a completely free market either.
The company is effectively cooperating with government bodies. That makes it difficult to predict what the next wave of AI releases will look like a year from now — especially because OpenAI’s products are used not only in the United States, but also in Europe, Asia, Africa, and other regions.
For other countries, this is also an AI national security problem. If a foreign AI product is used inside a country, it can potentially influence elections, political attitudes, the information environment, public health, education, data security, and critical systems.
That is why states will increasingly consider restrictions, controls, or localisation requirements for foreign AI products.
AI has long since become a strategic technology. Future model releases will not be discussed only as technological events. They will be governance events.
🟩 Competition Between AI Companies
There is another factor: competition.
While one company goes through checks, negotiates with the government, and delays its release, competitors may bring their models to market faster.
For example, Elon Musk’s xAI may use release speed as a competitive advantage. If OpenAI negotiates with the government for months, while another player completes the process in weeks, that player gains an advantage in publicity, users, and profit.
This creates a direct tension between AI safety and speed.
Powerful models must be tested. But the market does not like waiting. It pressures companies to release faster. If one company becomes too cautious, another can take its place.
That means AI regulation does not only affect safety. It also affects competition.
🟩 From Model Power to Governance Power
In the past, AI companies mainly competed over whose model was better. That is no longer enough.
The winner will not simply be the company that builds the most powerful model. The winner will be the company that can scale AI safely, quickly, and predictably.
Companies now need to improve not only the models themselves, but also the governance systems around them: access, control, security, monitoring, compliance, documentation, interaction with government, and risk management.
This is the new frontier of AI model governance.
The problem is that the regulatory framework either does not exist or becomes outdated almost immediately. And if lawmakers want to regulate AI effectively, they first need to understand the technology deeply.
Without that, regulation becomes formal, delayed, and weak.
🟩 What This Means for Business: AI Compliance Is Infrastructure
For business, this story means that AI tools can no longer be selected only by price or response quality.
Companies now need to ask harder questions.
What data is transferred to the model? Who has access to that data? How is it logged? Where is it stored? What restrictions does the provider impose? How often do the rules of use change? What guarantees does the provider give? Can the provider maintain access to the model after new negotiations with the government or regulator? How could future releases affect the company’s operations?
AI is becoming part of the infrastructure of almost every organisation.
Dependence on an external AI provider is no longer just a technical issue. It is a question of business resilience, AI compliance, security, and strategic planning.
This is where AI risk management becomes practical. It is no longer enough to ask whether a tool works. A company must understand how the tool is governed, what risks it creates, how those risks are monitored, and what happens when the provider changes the rules.
For businesses, AI control is now part of operational control.
🟩 The European Context: AI Policy Through Law
Europe follows a formal approach through the AI Act. This approach is built around documentation, transparency, risk assessment, cybersecurity, labelling, oversight, and the responsibility of providers, suppliers, and users.
It is a legalistic approach. Europe is trying to build a system of rules before the technology fully escapes control.
But the problem is obvious: AI develops too quickly. A law written today may already be insufficient tomorrow.
The United States and the European Union are taking different paths, but they are reaching the same general conclusion: powerful AI models can no longer be treated as ordinary software.
The United States is relying on flexibility, voluntary interaction with government, and national security. But this approach is too closed and too administrative.
The EU is relying on legal certainty, documentation, oversight, AI compliance, and responsibility. But this approach may be too slow for a technology that changes faster than legislation.
Neither the American nor the European model looks ideal.
In my view, the most logical solution would be an external independent AI regulator. Not just another bureaucratic office, but a technically competent institution capable of responding quickly to change, understanding real risks, and controlling the use of AI by both governments and private companies.
🟩 Main Conclusion
AI governance is no longer a theoretical topic. It is a practical necessity.
Organisations must control not only what AI generates. They must also consider risks connected with model access, data, future releases, dependence on providers, changing rules, and government intervention.
For companies, AI is a question of infrastructure, security, AI risk management, and predictability.
For governments, AI is a question of control, national security, and influence.
The main question is no longer which model is stronger.
The main question is who decides which model becomes available to the world, under what conditions, and how predictably that access will expand.
I will continue writing about this in plain language.
See you in the next piece.
Cheers,
