# Who Answers When an Agent Acts? By The Arrenite team | 2026-10-19 | 7 min read Series: The Human Equation ยท Issue 04 Canonical page: https://arrenite.com/letters/who-answers-when-an-agent-acts/ Written with AI assistance and edited by The Arrenite team. General reflection, offered as is, and not professional advice. The owner of a small hardware shop asked her purchasing agent to restock the screws. She meant forty boxes. The order that went out was for four thousand, from a supplier whose terms said no returns after seven days. Nobody noticed for a week. The agent had done exactly what it understood the task to be, and every step had looked reasonable. When the invoice arrived, her first words were not "I made a mistake." They were "The system did it." She said it for about a minute. Then she picked up the phone and called the supplier herself, because she knew, without anyone telling her, that the invoice had her name on it. This letter is about that minute, and about why it matters more with every month that AI agents get more capable. ## What the law says today We are not lawyers, and this is not legal advice. We can describe the direction of travel as of October 2026, and say where we could only confirm things second-hand. The common thread is simple: the responsibility stays with people and organisations. - **California.** Assembly Bill 316 became law in October 2025 and took effect on 1 January 2026. It says that a defendant who developed, modified or used AI cannot defend a harm claim by arguing that the AI acted autonomously. Other defences remain open, such as arguments about cause or foreseeability. (We confirmed the chapter number and dates from the official legislative record summary, but could not retrieve the chaptered text itself.) - **The European Union.** A revised Product Liability Directive treats software, including AI systems, as a product. Member states must put it into national law by 9 December 2026, and it applies to products placed on the market after that. Separately, the AI Act's rules for high-risk systems have been postponed, to December 2027 for stand-alone systems and August 2028 for those built into products. Reports we read describe the postponement as formally adopted in mid-2026, but sources differ on the publication details, so please check the Official Journal. - **The United States, federally.** In 2024 the Federal Trade Commission said there is "no AI exemption" from the laws already on the books, when it announced actions against companies using AI to deceive customers. That was a statement from that time and leadership, and we have not verified that it still stands in the same words. In February 2026 the National Institute of Standards and Technology launched an AI Agent Standards Initiative, aimed at standards for security, identity and interoperability. That is useful groundwork. It is not a liability rule. - **A small, telling case.** In 2024 a Canadian tribunal ruled against an airline whose website chatbot gave a customer wrong information about fares. The airline argued the chatbot was a separate thing responsible for its own words. The tribunal rejected that. It was a small-claims ruling, so it persuades more than it binds, but the reasoning is clear: you answer for what you put in front of your customers. Laws are still being written, and courts will draw many lines we cannot predict. Still, the direction is consistent. "The agent did it" is not a defence. ## Answerable is more than liable Liability is about who pays. Accountability is about something older. To be accountable is to be someone who can be asked, "Why did you do this?", and who can give an honest answer. It means being able to say "I was wrong," to make amends, and to be forgiven or to forgive. It is a relationship between persons. A shop owner can look the supplier in the eye. A system can produce an explanation, and some are very good at it, but an explanation is not the same as being answerable. A log of what happened does not apologise. A model that is retrained does not repent. This is why a fine and a lawsuit, though they matter, can never be the whole of it. Somebody has to be the one who stands there. Now picture a school administrator whose district has just switched on an agent that drafts and sends attendance letters to families. One morning, 300 families receive a letter saying their child is at risk of being dropped from the roll. It is a mistake, and several parents are frightened. The first question in the meeting is not "Which software?" It is "Who signs these?" If no one's name is on the process, then no one can pick up the phone and say, "We got this wrong, and here is what we are doing about it." That silence is the real harm, and it lasts longer than the error. ## What Scripture has to say The Bible does not mention agents, but it is deeply concerned with answering. In Genesis 3, after the first disobedience, God comes into the garden and calls out: "Where art thou?" (Genesis 3:9, KJV). It is not a request for information. It is an invitation to stand up and answer. What follows is a chain of passing the blame: the man points to the woman, the woman points to the serpent. It is the oldest "it wasn't me" in the book. In Genesis 4, Cain is asked about his brother and answers with a question: "Am I my brother's keeper?" (Genesis 4:9, KJV). The text treats the question as a dodge, not an answer. We are, in some real sense, answerable for what happens to others when it was in our power to prevent or care for it. Jesus puts the principle plainly: "For unto whomsoever much is given, of him shall be much required" (Luke 12:48, KJV). Power to act is a form of being entrusted. The more we can do, and the more we can delegate, the more is required of us. That is the old idea of stewardship: what is placed in our hands is not ours to hand off without care. An agent that can act at scale is a large entrustment. The Christian thread is not "never delegate." It is "know whose hands the thing is still in." ## An open question, an honest answer Could an AI system ever be accountable in its own right? Could it be asked "why?" and answer as a someone, be wronged or wrong others, and be forgiven? We do not know. That is the same open door we described in [Useful Is Not a Person](https://arrenite.com/letters/useful-is-not-a-person/) when we wrote about moral status, and we do not want to slam it. Serious people disagree. But whatever is true about the far future, today's answer is not open. No agent now running can be called to account in the way a person can. So accountability must stay with people and organisations, and it must be arranged on purpose, in advance, with names attached. "The agent did it" must never be an excuse, whether to a customer, a regulator, a family, or ourselves. This also protects us from a quieter temptation. If we let accountability drift into the machine, we may find we have let agency drift with it, and with it a little of what makes our work ours. > **Three small habits for the agent age** > > 1. Put a named human's name on every task you hand to an agent. If it goes wrong, that person is the one who answers, and who is able to put it right. > 2. Set a limit before you hand over the keys: a maximum spend, a list of things it may not send, and a point where it must stop and ask. > 3. Keep a short record of what you asked, so that when someone asks "why?", you can answer in your own words. ## Where this leaves us The shop owner never did get the screws sent back. She sold them over the following year, a few boxes at a time, to contractors who liked a good price. She also changed how she works: nothing over a set size goes out until she has read it herself, and the agent's work is checked each Friday. The tool is still useful. But the invoice still has her name on it, and she would not have it any other way. Being the one who answers is a weight, and it is also a form of dignity. --- *The Human Equation* looks at artificial intelligence, technological progress, economics and human meaning, and tries to separate what we can observe, what we can reasonably expect and what remains ours to decide. This letter is general information and is not legal advice. ## Sources and further reading - California Assembly Bill 316 (2025), Chapter 672, Statutes of 2025, adding Civil Code section 1714.46; summary from the California Legislative Information record and CalMatters Digital Democracy. Chaptered text not retrieved directly. - Directive (EU) 2024/2853 on liability for defective products (revised Product Liability Directive), Official Journal of the European Union. - Digital Omnibus on AI, delay of AI Act high-risk obligations (agreed May 2026, adopted mid-2026); known to us from law-firm summaries. Check the Official Journal for the final text and dates. - Federal Trade Commission, press release on Operation AI Comply (September 2024). - National Institute of Standards and Technology, "Announcing the AI Agent Standards Initiative for Interoperable and Secure Innovation" (17 February 2026). - Moffatt v. Air Canada, 2024 BCCRT 149 (Civil Resolution Tribunal of British Columbia, 14 February 2024). - Genesis 3:9, Genesis 4:9 and Luke 12:48, King James Version. Laws and plans are described as of October 2026 and may have changed. Please check the originals. *The Arrenite team* ## Terms in this letter - **AI agent**: An AI system that does more than answer: it can pursue a goal over several steps, use tools and services, and take actions such as booking, buying or sending, often with little supervision. - **accountability**: Being answerable for what you have done: able to be asked why, to give an honest account, to make amends. Different from liability, which is only about who pays. - **stewardship**: Caring responsibly for something that has been entrusted to you, such as time, money, creation or a gift, rather than owning it outright. - **moral status**: Whether a being matters for its own sake, so that we owe it something, such as not causing it needless harm. Whether and when AI systems could have it is an open question. - **agency**: The ability to make your own choices and act on them, rather than only having things done to you or for you.