We are being sold a comfortable story about artificial intelligence. The narrative goes like this: AI is powerful and occasionally dangerous, but technologists and regulators are aware of the risks. They are installing safeguards. The system is working, or at least it's trying to work. We should trust the process.
This framing deserves far more skepticism than it is receiving.
Consider the current discourse around AI safety. When prominent AI researchers warn that their own creations need "brake pedals," it sounds reassuring. The concern is being raised from inside the tent. Surely this means the industry is taking itself seriously. Surely this means we can relax.
But what does a brake pedal actually accomplish if no one is required to use it?
The pattern we are seeing is not new. It is the pattern of every transformative technology before its most consequential impacts became undeniable. The industry voices that speak about responsibility are given credit for responsibility, while actual regulatory frameworks lag years behind capability. The conversation shifts to how to safely deploy something, rather than whether it should be deployed at all, and at what pace.
This matters because the implied consensus around "responsible AI" masks a more troubling reality: we have no meaningful mechanism to enforce restraint.
When deepfake imagery of public figures emerges as a legal problem, when dating apps are explicitly designed to nudge users toward AI-mediated interactions, when workers discover their jobs have been automated without consultation, the guardrails we are told have been installed are revealed as largely ornamental. They are promises rather than rules. They are statements of intent rather than binding constraints.
The issue is not that technologists are callous. Many genuinely care about the implications of their work. The issue is that the economic incentives and competitive pressures in the technology sector are structurally misaligned with caution. If one company installs guardrails and a competitor does not, the competitor gains an advantage. If one country regulates AI development and another does not, the regulated country risks falling behind.
This is not a reason for defeatism. It is a reason for clarity about what the current conversation actually is.
When we hear that AI needs safeguards, we are hearing market participants advocate for the conditions under which they can continue scaling their products. That is not the same as submitting to genuine external oversight. It is not the same as accepting slowdown when slowdown is warranted.
The brake pedal framing also obscures a deeper question: Are we confident enough in our understanding of these systems to be deploying them at scale in the first place?
The honest answer is no. We are not. We are building and deploying systems whose failure modes we do not fully understand, whose long-term societal impacts we cannot predict, and whose capabilities are changing faster than our ability to study them.
The responsible path forward would involve more caution, slower deployment, and genuine regulatory teeth. It would involve asking whether a particular application of AI is necessary before rushing to build it. It would involve accepting that sometimes the right answer is to wait.
Instead, we are getting a consensus that frames this all as manageable. The brake pedal is being installed. The experts are aware. Trust the process.
This is the narrative we are being sold. And it is exactly the narrative that those who benefit from rapid deployment would sell.
The question is whether we are willing to think harder about what we are being asked to accept.