Financial Times columnist Robert Armstrong has argued, in a recent column, that despite the lofty promises AI executives make about ethical guidelines and safety, AI companies will ultimately behave like every other corporation: they will maximise returns to shareholders within whatever the law allows. When profit collides with internal safety principles, Armstrong writes, profit will win. His evidence is the cumulative weight of the numbers. Big Tech hyperscalers plan to invest over six hundred billion dollars in AI infrastructure this year alone. OpenAI raised one hundred and twenty-two billion dollars in a single recent funding round. AI startups raised seventy-three billion dollars in just the first quarter of 2025.
Capital that large brings expectations to match. Investors expect aggressive growth. Chief executives who slow down for safety reasons risk being fired. The argument is not that individual AI executives are insincere when they talk about safety in interviews. Many of them clearly care. The argument is that personal sincerity does not survive contact with structural pressure. When OpenAI reportedly missed user-growth targets last week, the Nasdaq dropped — a vivid demonstration of how investor pressure punishes any hint of slower expansion.
The historical analogues Armstrong reaches for are illuminating. Facebook executives in 2010 genuinely believed they were connecting the world. The company's advertising revenue model, however, pushed engagement algorithms toward outrage and addiction. Good intentions, bad incentives, predictable outcome. The 2008 financial crisis followed a similar pattern: bank executives knew mortgage-backed securities were risky, but bonus structures rewarded short-term deal volume. Individual ethics could not override systemic incentives. The point is not that humans are corrupt. It is that institutions reliably eat principle for breakfast.
Voluntary safety commitments — Anthropic's published guidelines, OpenAI's so-called Model Spec — are exactly the kind of corporate self-restraint that Armstrong's analysis predicts will fail when revenue is at stake. They have no legal force. They can be revised in a board meeting. The non-obvious point in his column is that the profit motive normally serves society well: it encourages innovation, lowers prices, expands access. The problem is specific to industries where externalities — harms imposed on third parties — are massive. Pharmaceuticals before regulation. Cars before seatbelts. AI today.
Armstrong proposes treating AI more like explosives — emphasising civil liability for harms rather than relying on a single sweeping law. Watch for fights over targeted rules: copyright, deepfakes, the use of AI in hiring, autonomous weapons. The historical parallel is the early twentieth century, when industries from automobiles to pharmaceuticals went from unregulated free-for-alls to carefully governed sectors. The transition usually only happened after disasters forced governments to act. The open question for the next five years is whether AI regulation arrives before or after its equivalent catastrophe. The answer will shape the labour market today's teenagers enter, the privacy they retain, and even what counts as truth online.
AI executives love talking about saving humanity from their own creations. But when profit collides with principle, capitalism has a 200-year track record — and humanity's safety usually loses.
Financial Times columnist Robert Armstrong argues that despite AI leaders' lofty promises about ethical guidelines and safety, AI companies will ultimately behave like any other corporation: they'll maximise shareholder returns within whatever the law allows. When profit conflicts with internal safety principles, he writes, profit will win every time.
He points to staggering numbers — Big Tech 'hyperscalers' plan to invest over $600bn this year, and OpenAI alone raised $122bn in a single funding round. Investors expect aggressive growth, and CEOs who slow down for safety reasons risk being fired. Armstrong concludes that meaningful AI safety must come from smart regulation, not corporate self-restraint.
The core idea here is that incentive structures beat good intentions almost every time. Don't focus on what executives say in interviews — look at who pays them and what those people want.
You're going to live and work in an AI-shaped economy — applying to college through AI-screened systems, competing for jobs against AI tools, and possibly building careers in fields that don't exist yet. How (or whether) governments regulate AI in the next five years will shape the labour market you enter, your privacy, and even what counts as 'truth' online. This isn't an abstract policy debate; it's the operating system of your adult life.
Armstrong proposes treating AI more like explosives — emphasising liability for harms rather than relying on a single sweeping law. Watch for fights over targeted rules: copyright, deepfakes, AI in hiring, autonomous weapons. The historical parallel is the early 20th century, when industries from cars to pharmaceuticals went from unregulated free-for-alls to carefully governed sectors — usually only after disasters forced governments to act. The question is whether AI regulation arrives before or after its equivalent catastrophe.