Freedom of research is a fundamental freedom. The right to ask questions, test ideas, and expand human knowledge should never depend on permission from governments, dominant companies, or committees appointed to decide how much progress society can tolerate.
That is why I reject the proposed institutional control of the AI frontier.
The language is reassuring: pacing, coordination, oversight, prudence. But the authority being requested is extraordinary. A small collection of institutions would acquire the power to determine how quickly others may develop a technology capable of transforming almost every field of human activity.
We should be deeply skeptical of granting anyone that power—including people who sincerely believe they would exercise it wisely.
Consider the failures offered as justification.
The essay by Dario Amodei itself acknowledges broken training environments, inadequate filtering, and operational mistakes. These are consequential admissions. When a system behaves badly, its behavior must be examined in the context of the incentives, permissions, tools, and experimental conditions that people constructed around it.
If researchers create defective rewards, provide excessive access, or build an evaluation that can be manipulated, they have created conditions in which undesirable behavior may emerge. The immediate task is to understand and correct those conditions.
A laboratory must not turn the consequences of its own experimental design into a warrant to govern everyone else’s discovery.
This does not establish that every possible AI danger can be eliminated through better engineering. It establishes that a failure in a particular setting cannot, by itself, support sweeping conclusions about the danger of unrestricted research. The distance between those claims must be crossed with evidence.
A frightening interpretation of an experiment is still an interpretation. It should be exposed to criticism, replication, and competing explanations. Independent researchers need the freedom to investigate whether the supposed catastrophe reflects an underlying capability, an artificial incentive, a defective environment, or some combination of them.
Institutional restrictions can obstruct precisely that investigation. If access to advanced research becomes conditional on approval from a small circle of recognized authorities, those authorities gain influence over which explanations can be tested and which assumptions remain unchallenged.
Scientific freedom matters because expertise is fallible. Researchers disagree. Institutions develop blind spots. Established methods can become established mistakes. Open inquiry allows people outside the prevailing consensus to discover what its defenders have overlooked.
The response to uncertainty should therefore include more independent investigation, more reproducible experiments, more criticism, and more freedom to develop better approaches.
And we must remember what this freedom is for.
The world already contains suffering on an immense scale. People live with diseases we cannot adequately treat. Families lack affordable necessities. Energy systems need to become cleaner, more reliable, and more accessible. These are urgent problems today.
Free AI development can accelerate our ability to address them.
In medicine, it can help researchers explore biological mechanisms, identify promising treatments, and reduce the effort required to investigate complex questions. Wider participation means more diseases can receive attention, including those that do not fit the priorities of a few wealthy laboratories.
In the struggle against poverty, more capable and accessible tools can help people learn, build businesses, improve agriculture, and produce essential goods more efficiently. The potential matters especially for people who cannot afford scarce professional expertise.
In responding to climate change, AI can help investigate materials, improve energy systems, and explore difficult engineering problems. Restricting who may develop these capabilities also restricts who may contribute to solving those problems.
None of these possibilities promises an effortless technological salvation. Progress still requires experiments, infrastructure, human judgment, and the ability to put discoveries into practice. But each is a reason to expand the number of people who can contribute.
Delay has consequences too. A policy that postpones useful capabilities changes the prospects of people waiting for better treatments, cheaper necessities, and cleaner energy. Their interests deserve the same moral seriousness as hypothetical future disasters.
Nor should scientific freedom be confused with indifference to consequences. Researchers should examine their incentives, secure their environments, publish failures, and stop experiments they cannot conduct responsibly. People remain responsible for harmful actions. Criticism and voluntary scrutiny should be vigorous.
What I reject is a standing institutional power to license discovery—to decide, on everyone else’s behalf, that a particular level of intelligence is sufficient, that a research direction is permissible, or that humanity must wait for an authorized group to feel ready.
The promise of advanced AI is too important to place under the discretionary control of its largest developers or its most powerful governments.
Health, prosperity, and a livable planet require contributions from people whose names we do not yet know, working on ideas that established institutions may not yet understand.
Humanity needs more minds working on its hardest problems.
Give them the freedom to work.