This article was originally published in Finextra on 26 June 2026.
Yuval Noah Harari’s warning in the Financial Times about granting legal personhood to AI agents should be read carefully by those of us in financial markets. His concern is not science fiction. It is about accountability: what happens when a non-human actor can own assets, make decisions, pursue profit and exploit loopholes, yet remain beyond the reach of the sanctions that discipline human behaviour?
For markets, that question is not abstract. It cuts to the heart of trust.
Prior to the pandemic – and before the launch of ChatGPT in 2022 – I argued that AI and robots should not be granted legal personhood, particularly in finance. My reasoning was simple: financial markets depend on responsibility. If an AI trading engine violates rules, manipulates markets or causes harm, the question cannot be allowed to become what the machine intended. It must be which human firm deployed it, supervised it, profited from it and failed to control it.
Harari brings this concern into sharper focus. He asks what kind of sanctions could keep a non-human corporation in check. The same question applies to a non-human trading engine. You cannot imprison an algorithm, shame it, or appeal to its conscience. If the worst consequence is deletion or replacement, the incentive structure is fundamentally different from the one that governs human traders, managers and directors.
That matters because markets are not merely mathematical systems. They are human institutions: built on rules and incentives, sustained by enforceable accountability, and ultimately dependent on trust. Modern market microstructure already contains countless vulnerabilities: fleeting liquidity, fragmented venues, latency gaps, queue positioning, predictable order-routing, and the mechanical reactions of slower participants. A sufficiently powerful AI system, operating at scale, could discover and exploit these patterns far faster than any human.
Some will call this efficiency. Sometimes it may be. But there is a difference between improving price discovery and industrializing predation.
A human trader who repeatedly exploits a market flaw can be questioned, investigated, disciplined or prosecuted. A firm that encourages such behaviour can be sanctioned. But if a trading engine learns that manipulative patterns are profitable, whether spoofing-like behaviour, liquidity games, momentum ignition, or strategies that prey on predictable retail flows, who is responsible? The coder who built the model? The compliance officer who failed to understand it? The firm that plugged it into live markets? My answer has not changed: the buck must stop with the firm that deploys the AI.
In 2020, I wrote in the Financial Times that regulators must get into the sandbox with AI innovators earlier in the development cycle. The point was not to smother innovation. It was to prevent regulators being introduced to powerful systems only at deployment, when commercial incentives are already entrenched and the technical complexity is already beyond easy inspection.
That argument now looks more urgent. AI trading engines should not be treated as black boxes whose decisions are beyond comprehension and therefore beyond liability. If a system is too complex for a firm to explain, supervise or control, it should not be operating in live markets with other people’s money.
This is not an anti-technology argument. Technology has transformed finance for the better, reducing costs, improving access and widening participation. But finance has lived through enough crises to know that innovation without accountability often ends badly. The lesson of 2008 was not that complexity is inherently bad. It was that complexity becomes dangerous when those who profit from it can avoid responsibility for its consequences.
The same principle applies to AI. The more powerful the machine, the greater the responsibility of the humans behind it.
The practical implications are clear. AI systems used in markets should be auditable. Their objectives and constraints should be documented. Their behaviour should be monitored in real time. Firms should be able to explain the governance framework, risk limits and escalation procedures under which a system operates. Regulators should work with innovators early, not arrive after the damage is done. And legal responsibility should remain firmly with the licensed, capitalised, human-led institutions that put these systems into the market.
A further question, which will become increasingly important as retail platforms integrate AI into trading tools, is whether markets need a clearer distinction between genuinely human-generated retail flow and AI-assisted retail activity. That may ultimately require its own framework, and its own debate.
The alternative is a market in which profit-seeking machines exploit human systems while humans deny responsibility for what those machines do. That would not be progress. It would be the erosion of the very trust on which markets depend.
AI can assist traders, improve analysis and strengthen compliance. It may even make markets more efficient. But it must never become a legal shield. The future of finance should not be one in which machines are treated as persons. It should be one in which people remain responsible for the machines they unleash.
