AI companies are just companies — Robert Armstrong on the limits of corporate self-regulation
Source · Technology desk
— Summary
In an Unhedged column, the FT's Robert Armstrong argues that AI firms will follow the same one rule as any other corporation — maximise shareholder return up to the limit of the law — and that hopes of self-regulation in AI safety are misplaced. He contrasts the optimist analogy (workers freed from horse-drawn carts) with the doomer retort that this time we're the horses, and notes the equine population's fate in the early 20th century. The argument: when corporate principles conflict with profit, profit wins.
The numbers underline the pressure. Big Tech "hyperscalers" plan to invest **more than $600bn** in AI in 2025; AI start-ups raised **$73bn in Q1 2025**; OpenAI raised **$122bn in a single funding round** last month. That capital, Armstrong writes, comes from investors who demand high returns and know more capital will be needed for compute — making the industry "extremely sensitive" to revenue growth. He cites the Wall Street Journal report this week that OpenAI missed internal sales and user targets, which moved the entire Nasdaq before OpenAI dismissed the report as "clickbait".
Anthropic CEO Dario Amodei is quoted on the tension between not "autonomously threatening humanity" and staying ahead of authoritarian rivals; Armstrong's position is that this tension is "noise" relative to the profit incentive. His regulatory prescription is plain: don't try to protect specific job categories; match tools to specific harms (physical, digital, psychological, financial); rethink agency law for non-human agents; and emphasise liability over duty-to-warn so investors have skin in the safety game. AI is new, capitalism is not. Source: Financial Times, 30 April 2026, Robert Armstrong.
AI companies are just companies
The story in one line: Robert Armstrong’s Unhedged column argues AI firms will obey profit before safety like any company, that the capital math (>$600bn 2025 hyperscaler capex, OpenAI’s $122bn round) makes self-regulation unrealistic, and that liability — not duty-to-warn — should be the centre of AI policy.
Why it matters
Two ideas sit behind the column. First, the analogy fight: optimists say AI workers will reskill the way buggy-whip makers became autoworkers; doomers say in this technology, humans are not the drivers but the horses (whose population collapsed once cars arrived). Second, the capital reality: investors who put $122bn into a single OpenAI round are not patient with safety detours that lower revenue growth — they will replace any CEO who chooses safety over speed.
Armstrong is careful not to dismiss safety pledges as cynical. He says he believes Dario Amodei is sincere when Amodei talks about steering AI “away from negative outcomes”. His point is that the relevant incentive structures don’t care what Amodei is focused on. From there, the policy follows: tools matched to harms, liability put on investors, and rejection of the framing that AI is too complicated for citizens to legislate.
Takeaway
For investors, the column is a reminder that the AI capex super-cycle and an AI safety regulatory super-cycle are now both running. For executives at OpenAI, Anthropic and the hyperscalers, it formalises the criticism that any deceleration on safety grounds will be punished by capital. Watch the next regulatory move — the column reads as a soft brief for a US AI liability statute.
Source: Financial Times, 30 April 2026, Robert Armstrong.