Who answers for an artifact that acts

For the whole history of tools, responsibility had nowhere to hide. The hammer holds no ambition for the nail; whatever the tool did, you did. Agentic systems break that comfort, because they generate conduct nobody scripted, and into the gap steps a familiar chorus. The builder says the deployer configured it. The deployer says the model misbehaved. The vendor’s terms say the output is the user’s responsibility. Everyone in reach has a plausible story, and a harm with only plausible stories around it is a harm nobody pays for.

The first courts to face the question have shown little patience for the chorus. When Air Canada’s website chatbot invented a bereavement-fare policy and a grieving customer relied on it, the airline argued, remarkably, that the chatbot was a separate legal entity responsible for its own statements. The British Columbia tribunal rejected the argument and held the airline liable for what its artifact had said. The ruling is small, a consumer tribunal and a few hundred dollars, and its logic is the important part. The company chose to put the actor in front of the public, presented it as speaking for the company, and profited from its operation; therefore the company answers for its conduct. That is not a new theory of machine minds. It is the tool-user dyad restored by force: if you will not supervise your actor, the law will treat its acts as yours.

Expect that logic to become the default, because every alternative is worse. Assigning liability to the artifact itself creates an uncollectable defendant. Assigning it to users penalizes the party with the least visibility into the system. Assigning it to nobody subsidizes recklessness at scale. Deployer responsibility is the only address where accountability, control, and benefit already live together, which is the test liability law has used since before software existed.

For anyone operating agents, the move is to get ahead of the default instead of discovering it in a ruling. Deploy every agent as if its conduct were your signature. Bound what it can do rather than what it is told to do, log what it did, and staff the answering. The threshold rule says when a system has become an actor; this brick says what that costs, and who is billed. The bill is also the blueprint. An industry that knows the deployer answers is an industry with every incentive to bound, log, and staff its actors well, and accountability priced early is the cheapest safety anyone will ever buy.