The Fallacy of Voluntary Compliance
Andrew Bailey, Governor of the Bank of England, recently suggested that regulation might not be the right starting point for managing artificial intelligence. The argument hinges on a desire to foster innovation and prevent the stifling of progress before we even understand the full scope of the technology. Instead, proponents of this view often call for rigorous testing and voluntary commitments. However, asking a corporation to police itself is like asking a wolf to guard the hen house. Currently, the industry relies heavily on "morally binding" agreements—promises made by the very entities that stand to gain billions from the rapid deployment of these systems. These agreements lack teeth. They are rhetorical tools used to buy time, allowing companies to outpace the slow machinery of government oversight.
This "testing-first, regulation-later" framework treats AI like a new chemical compound in a lab, but AI behaves more like a social force. By the time a model passes a series of internal benchmarks, its influence may have already permeated critical infrastructure. Relying on self-reported safety data creates a vacuum of accountability. If a model fails or causes systemic harm, the company can point to its tests as proof of due diligence, even if those tests were designed by the same engineers who prioritized speed over security.
The Hidden Cost of the AI Race
Political leaders often frame the development of AI as an arms race, a geopolitical necessity where being second means being obsolete. This narrative creates a sense of urgency that justifies cutting corners. In this frantic sprint toward AGI (Artificial General Intelligence), the conversation stays stuck on "frontier risks"—catastrophic, hypothetical scenarios like rogue superintelligence or the synthesis of novel pathogens. While the threat of bioweapons is a valid concern for national security agencies, focusing solely on these existential dread scenarios serves a specific purpose: it distracts from the harm occurring right now.
When we talk only about the apocalypse, we stop looking at the current victims. This creates a hierarchy of risk where the theoretical death of millions in a sci-fi future carries more weight in policy discussions than the very real, documented harms occurring in our own neighborhoods. The race to win the AI crown treats progress as an absolute good, assuming that any technology that advances the state of the art is inherently worth the risk, regardless of who gets crushed in the process.
The Slow Violence of Algorithmic Implementation
While the tech giants debate "containment" and "alignment," a different kind of damage is being done through what scholars call "slow violence." This is the gradual, often invisible destruction caused by technological integration. It shows up in healthcare algorithms that systematically underestimate the needs of Black patients, leading to unequal access to life-saving care. It manifests in the massive energy consumption required to train LLMs (Large Language Models), which drains local grids and exacerbates environmental degradation in regions least responsible for carbon emissions. It appears in the massive displacement of labor in the Global South, where workers are hired for low wages to label data in precarious, psychologically taxing conditions.
These aren't bugs in the system; they are features of how AI is being deployed. The focus on "closed-loop" models ignores the fact that AI is already an open wound in many communities. For a person whose credit score is ruined by a biased credit-scoring algorithm, or a gig worker whose livelihood is dictated by an opaque, unchallengeable app, the debate over whether AI might one day build a bomb is a luxury they cannot afford. The immediate humanitarian impact is being treated as a secondary concern, a problem to be solved once the technology is "safe."
Historical Echoes and the Power Struggle
This tension between innovation and oversight is not new. During the Industrial Revolution, factory owners argued that early labor laws would destroy economic competitiveness and stifle the ingenuity of the age. They successfully used the promise of future prosperity to delay safety standards, child labor prohibitions, and wage protections for decades. The pattern is repeating. Today, the argument that regulation is a burden on innovation is the modern version of the old defense of unbridled industrialism.
Ultimately, the debate over AI governance is not a technical puzzle to be solved by engineers; it is a power struggle. It is a fight over who gets to define what "safety" actually means. Is safety the absence of catastrophic biothreats for a handful of superpower nations? Or is safety the protection of privacy, the preservation of fair labor, and the prevention of systemic bias for the global population? If we allow the technology to outpace our ability to govern it, we are effectively handing the keys to the future to those who profit most from its chaos. The question is not whether we can afford to regulate, but whether we can afford to let the definition of safety be written by the corporations themselves.
