The Fallacy of Testing-First
Andrew Bailey, Governor of the Bank of England, recently argued that regulating AI is "not the right place to start." His preference for a testing-centric approach suggests that we should focus on technical benchmarks and red-teaming before jumping into heavy-handed legislative frameworks. This perspective relies on a specific brand of technocratic optimism. It assumes that if we can identify a technical glitch or a model failure in a controlled environment, we can patch it like software before it causes systemic harm. But this mindset treats AI risk as a purely engineering problem, akin to fixing a leaky pipe in a bank's server room.
Technological safeguards do little to address the "slow violence" inflicted by algorithmic systems. When an AI model exhibits bias in credit scoring or healthcare triage, the harm is not a sudden explosion; it is a quiet, continuous erosion of opportunity. A laboratory test might catch a prompt-injection vulnerability, but it rarely captures how a predictive policing tool disproportionately targets specific ZIP codes or how automated housing assessments strip dignity from low-income tenants. These are not technical bugs to be patched in a "closed loop" system; they are fundamental societal impacts that testing alone cannot measure.
By focusing on "frontier AI"—the massive, next-generation models—Bailey and his contemporaries shift attention away from the AI already in use. While experts debate the existential risks of future superintelligence, existing machine learning (ML) models are already making life-altering decisions. People in marginalized communities encounter these models every day through automated welfare eligibility, migration screening, and medical diagnostics. Focusing only on the most powerful models ignores the immediate, tangible damage caused by the models we have already deployed.
The Mirage of Voluntary Compliance
Recent discussions between government officials and tech giants have centered on the idea of "morally binding" voluntary commitments. This concept sounds noble in a press release, yet it lacks any mechanism for enforcement. History shows that when corporations are asked to self-regulate, they often use the process to secure a seat at the table while delaying actual oversight. These voluntary frameworks act as a strategic buffer, buying time for companies to solidify their market dominance before democratic institutions can pass enforceable laws.
A
