Many Hands

If an AI agent does something in your name and it goes wrong, who answers for it?

Work that once passed from one person to the next now passes through a model built by one company, a platform run by another, an integration written by a third, the employee who set it up, and the manager who approved the result. Each of those hands contributed, and each has a reasonable explanation. Political philosophers call this the problem of many hands: when many people contribute to an outcome, it becomes hard to hold anyone responsible for it.

Many Hands separates three questions that used to have one answer. Who pays? Who has to explain what happened and fix it? Who should apologize? Agent systems pull these apart, and the blame tends to land on the nearest person, such as the one who signed or approved, rather than on the decisions that shaped the system.

The book follows responsibility through cases that have already been decided, from a self-driving test car in Arizona to a benefits scandal in the Netherlands, a chatbot's promise in Canada and fabricated case citations in a London court. It then turns to Europe's new and staggered rules: what the EU AI Act asks of the people who oversee high-risk systems and when those duties apply, what changes when the new EU Product Liability Directive treats software as a product, and why that directive covers injuries and damage to private property and data but not pure financial loss, the wrong payment or the bad contract where many agent mistakes land. It follows the money down the vendor chain, through liability caps and insurers, and into the workplace, with what German law says about works councils and about how much an employee has to pay when something goes wrong.

The last part asks when it is fair to hold the person who signs responsible, and what organizations can build so that someone can still answer: a named owner for each agent, authority limited to the task, records of who decided what and on what authority, ways for affected people to appeal, and limits on how those records may be used. An appendix offers two short lists of questions to ask before an agent acts in your name, one for individuals and employees and one for organizations.

Written for European readers, especially managers, works council members, freelancers and small-business owners, Many Hands needs no legal or technical background. Every legal and factual claim was checked against its primary source, and the notes cite those sources with dates and links. Where a company or person is criticized, their response or the status of the case appears next to the criticism. The law is stated as of 16 October 2026. The book is not legal advice.

The book is open about how it was made: its chapters were drafted by AI agents under the author's direction, reviewed independently by AI reviewers, checked against primary sources, and selected, edited and approved by the author.