With the requirement of prior approval for new models, AI regulation is no longer about safety and has become the ultimate competitive barrier.
There is a curious detail in the launch of GPT-5.6, whose limited preview was announced to a select group of partners: OpenAI is not just presenting variations of power and cost. According to Engadget, the company divided the model into three variants, including the most powerful and the most affordable to date. The portfolio strategy is obvious, but what really matters here is the architecture of power taking shape around these releases. We are witnessing the end of Silicon Valley's old "move fast and break things" mantra — and not out of a crisis of conscience among engineers, but because the State has finally set up the tollbooth.
The growing demand that advanced models undergo case-by-case government approval is no mere bureaucratic tweak. It is the end of total freedom to launch. By turning regulation into a de facto licensing regime, governments have created a bottleneck where there used to be an open road. If you need a government stamp before training or releasing a large-scale neural network, you are not operating in the free market; you are operating under a concession.
In my view, this establishes a new paradigm where regulatory compliance becomes the technology's greatest competitive barrier. OpenAI, with its army of lawyers and robust safety structure, can navigate this bureaucratic swamp and negotiate previews with oversight entities. But a brilliant startup with two million dollars in seed money and a revolutionary architecture idea? It dies on the vine before it can even compile its first model. Bureaucracy acts as an invisible moat, protecting incumbents from any real bottom-up threat.
The finest irony of this story is that the very companies that loudly demanded regulation over the past two years — in the name of preventing the arrival of "unchecked AI" — are its greatest beneficiaries. By embracing the apocalyptic narrative, they helped design a licensing system where only they have the capital and the structure to pass the bar exam. The existential risk of artificial intelligence, we now discover, was less about robots taking over the planet and more about the consolidation of an oligopoly with the State's blessing.
GPT-5.6 and its three variants are, therefore, a symptom of a market that has aged decades in a matter of months. AI is no longer the Wild West of 2022. It is now a regulated sector, where innovation needs to get its papers in order and toe the line. And as the government turns into the final arbiter of what can or cannot be intelligent, the lingering question is disturbing: are we protecting society from dangerous machines, or just protecting the biggest AI companies from their own competition?
By requiring case-by-case government approval to train or release large-scale models, regulation acts as a licensing regime. Incumbents with massive legal resources can navigate this bureaucracy, while underfunded startups die before compiling their first model.
By promoting the narrative of existential AI risk, major tech companies helped design a licensing system that only they have the capital and structure to pass. This turns regulation into an invisible moat that consolidates their market dominance into a state-blessed oligopoly.
The free-market era of rapid, unchecked AI launches is over. The AI sector is now a regulated industry where compliance is the ultimate competitive barrier, and governments act as the final arbititer of what models can be released.