The Myth of Self-Policing
Andrew Bailey, head of the Bank of England, recently suggested that heavy-handed regulation might not be the best place to start with artificial intelligence. This sentiment echoes through Silicon Valley, where tech executives argue for a "testing-first" model. They suggest that we should observe how models behave in controlled environments before we codify laws to restrict them. It sounds reasonable. It sounds cautious. Yet, this approach relies on the shaky premise that corporations can effectively police their own products while simultaneously chasing astronomical returns for shareholders.
The tension here isn't just about the speed of innovation; it is about a fundamental conflict of interest. When companies propose "morally binding" agreements to ensure safety, they are essentially asking the public to trust their integrity. History shows that when profit motives collide with public safety, profit almost always wins. A company might spend millions on red-teaming a Large Language Model (LLM) to prevent it from teaching someone how to make a pathogen, but that same company has little financial incentive to fix the subtle, systemic biases that drive away customers or create legal liabilities in smaller, less profitable markets.
If the goal is to win an "AI race," safety becomes a secondary variable. In a geopolitical sprint, being second means being irrelevant. This pressure creates a race to the bottom where the first entity to release a functional, even if slightly unstable, tool captures the market. Waiting for regulation to catch up to these breakthroughs is like trying to legislate the speed limit after the cars have already crossed the border.
Diversionary Tactics and Frontier Risks
Current discourse often centers on "frontier AI" risks. We hear warnings about autonomous weapons, bio-threats, and the potential for a sudden intelligence explosion. These are valid existential concerns, but they serve a specific political function: they shift the focus away from the harms happening right now. By talking about the end of humanity, we avoid talking about the erosion of human dignity in the present.
While experts debate the probability of an AI-driven apocalypse, real people are already losing homes because an opaque algorithm miscalculated their creditworthiness. In many cities, predictive policing tools—often criticized for their feedback loops—are used to justify heavy-handed law enforcement in specific neighborhoods. These are not theoretical risks. They are current, measurable consequences of unregulated deployment. By framing the debate around high-concept sci-fi scenarios, the industry effectively bypasses the need to address immediate socio-economic fallout.
The delay in regulation creates a vacuum. In this vacuum, the harms of algorithmic bias in healthcare diagnostics and housing assessments are treated as "unfortunate bugs" rather than systemic violations of rights. If we wait for the "frontier" to arrive to start regulating, we have already allowed a new architecture of inequality to settle into the bedrock of our social institutions.
The Ghost of the Industrial Revolution
The "wait and see" approach is not a new phenomenon. During the early Industrial Revolution, the sudden introduction of chemical processes and unregulated factory labor caused widespread social upheaval and environmental degradation. We did not wait for the smog to become unbreathable or for the rivers to become toxic before deciding that industry needed oversight. We developed labor laws, environmental protections, and safety standards in response to visible, documented suffering.
The distinction today is that AI is software, which feels intangible and therefore less dangerous. But the impact is physical. It translates into economic displacement for workers in the Global South, where labor arbitrage is being replaced by automated service bots. It translates into the loss of cognitive autonomy. Unlike the steam engine, which required physical coal and iron, AI scales at the speed of light. We do not have the luxury of a slow, reactive regulatory evolution. The damage from an unconstrained, biased, or discriminatory algorithm can be replicated a billion times over in a single afternoon.
Who Does the Delay Protect?
We must ask a blunt question: who benefits from this delay? It is not the person being denied a loan by a faulty algorithm. It is not the displaced worker or the community targeted by biased policing. The beneficiaries are the entities that hold the intellectual property and the capital. A delay in regulation is a subsidy for rapid deployment. It allows companies to extract value from new technologies while pushing the costs of errors, biases, and social instability onto the public.
The argument that regulation stifles innovation is a false binary. Some of the most successful industries in human history—aviation, pharmaceuticals, nuclear energy—exist within tight regulatory frameworks. People fly because they trust the FAA. People take medication because they trust the FDA. In these sectors, regulation is not a barrier to progress; it is the foundation of public confidence. Without that confidence, the social license to operate will eventually vanish.
Relying on voluntary commitments and "testing" is a tactical diversion. It offers the illusion of oversight without the teeth of accountability. If we want to ensure that AI serves humanity rather than just optimizing for a specific set of corporate metrics, we cannot afford to wait for the breakthrough. We need rules that protect human dignity today, before the systems of tomorrow have already decided our fate.
