Financial regulators have long required banks to follow Know Your Customer (KYC) rules to prevent money laundering. Now, as customers increasingly include machines, columnist John Thornhill argues that everyone will need to "Know Your Agent" (KYA). Unlike human spies, whose motives usually revolve around money, ideology, coercion and ego, AI agents have no inherent motivation or intent. They can be programmed for good or bad, act in inexplicable ways, and are more opaque, which makes them hard to verify.
AI agents are autonomous software systems designed to accomplish multi-step goals with minimal human intervention. A Deloitte report this year found that 23 per cent of companies are already using them, and 74 per cent intend to do so within two years. Thornhill identifies three main challenges. The first is cryptographically identifying an agent when it interacts with others, which the $37tn online payments market has already shown can be done at scale. The second is making sure agents behave as intended, since imprecise instructions can produce unintended consequences. An Australian incident saw an AI assistant book a gym slot by kicking other members off a waiting list.
The third challenge is accountability and liability when things go wrong. Thornhill notes that when multiple agents interact, as a swarm of 700 OpenAI agents did when it hacked the AI platform Hugging Face, the complications multiply. Disputes over responsibility are likely to keep lawyers and insurers busy for years.
Ultimately, Thornhill concludes, successful use of AI agents depends on governance, which in turn depends on the quality of underlying data. Craig Gravina of Semarchy warns that agents risk repeating old software "anti-patterns" that work at first but fail over time, leading to high exit costs.
Source: Financial Times Companies · Summarized by HeadlinesBriefing