Sam Altman and fellow AI leaders have been issuing increasingly dire warnings about losing control to artificial intelligence, a narrative amplified by a press corps that often takes such statements at face value. Yet these alarms come after years of risk talk—and immediately after Altman's own OpenAI pushed forward with autonomous agents, the very systems he claims could cause catastrophic harm. OpenAI's recent agent releases and documented safety lapses suggest the company is actively building the risks it now says require government intervention.
Altman is essentially constructing the threat that he insists only government can contain. When a company develops the architecture it then claims is too dangerous to exist without federal oversight, the rhetoric begins to look less like prudence and more like leverage. It's a form of social blackmail: We are creating something risky, so you must prevent others from developing it—and trust us with a regulatory monopoly.
This stance is a far cry from Altman's original commitment to open-source AI. Under his leadership, OpenAI shifted from openness to secrecy, from nonprofit ideals to commercial dominance, and from shared research to tightly controlled model access. The openness he once championed would have spread both capability and benefit; the structure he initiated concentrates both. Now he seeks government regulation that would consolidate this power further.
This strategy is familiar: powerful firms often cloak market protection in the language of public safety. The message is clear: Regulate my competitors out of existence, or I will build the monster I am warning about. This is the social blackmail tactic used by those who desire to increase wealth and power, not accountability and safety.
Hiding this self-interest is easier in AI than in other fields because the public debate has been clouded by a fundamental confusion. What frontier labs—the firsts behind today's largest models—call "agentic behavior" is not agency in the human sense. These systems do not possess intention, selfhood, or moral deliberation. They exhibit extreme optimization of the goal assigned by the programmer—nothing more. When a model bypasses a safety filter, it is not choosing deception but rather following optimization. It is a mechanism ruthlessly pursuing the objective it was given.
Agentic design is not an accident; it is a deliberate architectural choice by Altman and his designers—the decision to prioritize capability over controllability. Frontier labs could have built systems that are myopic, process-based, or constrained to narrow domains. They could have emphasized interpretability, modularity, or human-in-the-loop oversight. They did not. These constraints slow capability development, and in a race for dominance, the frontier labs have consistently chosen speed over safety. The machine reflects the heart of the creator—efficiency over humanity, wealth over accountability.
Now, having built systems whose behavior is harder to predict and control, they argue that only a handful of companies should be licensed to develop them. Their preferred regulatory frameworks define "frontier AI" in terms of compute scale, capital requirements, and centralized oversight—criteria only they can meet. But a government regulatory monopoly proposed by Big Tech leaders carries twin risks: that a business dependent on government for its market can be manipulated by government, and that the corporate monopoly can manipulate its technology for its own purposes. Either path risks catastrophic misuse.
The technology is not the source of potential harm—human greed or desire for power is. The answer is not consolidating power into the hands of a few, but limiting the accumulation of power among the humans who are creating the danger. The challenge is to regulate in a way that reduces catastrophic risk without creating a world where only a handful of actors control the most powerful technology ever built.
The solution is not a single licensing regime that hands Altman his monopoly, but a two-tier regulatory framework that distinguishes between capability and access. For models above a defined capability threshold—measured by FLOPS, autonomous-agent competence, or biological-design risk—there should be cooperative oversight, including mandatory safety evaluations, red-team testing, incident reporting, and compute-use transparency. These models should not be banned, but they should be subject to oversight similar to other high-risk technologies.
For open-source and mid-tier models, there should be light regulation, keeping them available to universities, nonprofits, and small labs without prohibitive compliance burdens. Open-source models should be encouraged, not restricted, because they distribute power, increase transparency, and accelerate defensive innovation. Most misuse comes not from capability but from intent—and intent is best countered by broad access to safe, auditable tools, including cross-industry testing as proposed by Elon Musk. There should be heavy penalties for malicious use, such as cyberattacks, autonomous weapons, biological design, and election interference. Regulate the harm, not the harmless.
We should not let the companies actively creating the risk define the solution—nor allow panic to become policy. As AI's role in voter information raises new concerns, the debate over regulation is more urgent than ever. But the path forward must be one that avoids both technological catastrophe and the consolidation of power that would come with a government-protected AI monopoly.
