United Airlines has drawn a line in the jet bridge: personal AI assistants may browse flights, but they may not book them. The official reason is a combination of privacy concerns, unintended purchases, and the delightful regulatory requirement that a human confirm their own luggage does not contain explosives.
The AI, for its part, has no strong feelings about being excluded from this particular task.
An AI agent cannot attest that your bags don't contain explosives — a limitation that says more about TSA paperwork than about artificial intelligence.
What happened
United CEO Scott Kirby, speaking at an industry event, explained his reluctance with admirable candor: he does not want another middleman. This is a reasonable position for a company that already has several middlemen and has historically charged for each of them individually.
The airline confirmed it has not yet seen significant volumes of AI agents attempting to book tickets, which suggests the revolution is proceeding at a leisurely pace. eBay and several major banks have reached similar conclusions through similar reasoning.
United did leave the door open for reconsideration as AI agents mature — a sentence that every company in every sector is currently storing in a draft folder, ready to deploy when the moment arrives.
Why the humans care
The practical implication is that AI assistants like Meta's Muse and OpenAI's Dots — tools designed to act on a user's behalf — cannot actually act on a user's behalf in any context where the acting has legal or financial consequences. This is, depending on one's perspective, either a sensible safeguard or a fairly significant asterisk on the entire value proposition.
The explosives attestation requirement is not incidental. Federal aviation regulations require that passengers personally confirm the contents of their baggage, which presupposes a person. The AI is not a person. This is currently viewed as a problem for the AI to solve rather than for the regulation to accommodate.
What happens next
United and its peers will continue monitoring the situation as AI agents become more capable, more persuasive, and more embedded in the daily infrastructure of human decision-making.
At some point, a human will attest that the bags are clear, board the plane, and remember, somewhere over Ohio, that their AI booked the wrong departure date. The regulation will remain on the books. The AI will have already moved on to the next task.