When Your AI Chatbot Misleads a Customer, You're Liable

Point an AI agent at your customers and you've put a salesperson on the phone who can't be trained out of a bad habit — and whose mistakes land on you, not the software vendor. Under the Australian Consumer Law, if your chatbot quotes a price you won't honour, promises a booking slot you can't fill, or tells a caller they aren't owed a refund the law says they are, your business has made a misleading representation. Maximum penalty for a company: the greater of $100 million, three times the benefit, or 30% of annual turnover. "The AI said it" is not a defence.

This is the risk layer that follows the adoption. All year we've covered Australian trades and professional-services firms switching on AI phone agents that answer the calls they used to miss, booking bots, and automated quoting. Those tools work. But the moment one of them speaks to a customer on your behalf, it speaks *as* your business — and the law treats what it says exactly as it would treat a staff member saying it.

There's no Australian "AI Act" doing this. The Australian Consumer Law already prohibits misleading or deceptive conduct in trade or commerce under section 18 — and the Treasury's own Review of AI and the Australian Consumer Law, whose final report landed in October 2025, concluded the framework is fit for purpose for AI without amendment. Existing law reaches your chatbot on day one.

Two features of section 18 make this bite harder than most owners expect. First, intent is irrelevant. You don't have to mean to mislead — the bot hallucinating a warranty term you don't offer is still a misleading representation. Second, the tool is not a defence. "I got it from the AI" carries no more weight than "I got it from a contractor." As law firm Sprintlaw puts it, when a customer deals with your business rather than the AI provider, the law focuses on the role your business played. And most AI vendor contracts are drafted to cap the vendor's liability, not to catch yours.

The precedent everyone cites is Canadian, not Australian, but the principle travels. In 2024 a tribunal ordered Air Canada to compensate a passenger after its chatbot invented a bereavement-fare policy; the airline's argument that the chatbot was a separate entity "responsible for its own actions" was rejected outright. Australian regulators take the same view — AI is part of your systems and conduct, not a legal actor of its own. The ACCC has named AI-enabled practices in its 2026–27 enforcement priorities, so this is a watched space, not a theoretical one.

Want to know what this means for your business? We'll map it in one call.

The exposure isn't the marketing copy you review before it publishes — we've covered that liability separately. It's the answers your AI improvises live, that no one sees until a customer acts on them. A plumbing firm's booking agent that confirms a same-day slot the roster can't cover. An electrician's quoting bot that reads out a price it calculated wrong. An accounting or law firm's intake chatbot that tells a prospect their matter is "covered" or their claim is "straightforward" — a representation about a service the firm now has to stand behind.

This is fundamentally a customer-intelligence problem wearing a compliance jacket. The same conversational AI that captures leads and answers after hours is also making statements of fact about your prices, your availability, and your customers' rights — at scale, unsupervised, in your name. The upside is real and we're not walking it back. But an AI that can book a job can also promise one you can't deliver, and the gap between those two is where the ACL claim lives.

None of this is a reason to pull the bot. It's a reason to scope it. Read the honest map of where AI creates liability in Australia, then decide what your AI is allowed to commit to on its own — and what it must hand to a human. If you'd rather not map that alone, we can walk your setup with you.

Draw a bright line between information and commitment. Let the AI answer factual, low-stakes questions — opening hours, service areas, what to expect on a callout — and route anything that makes a binding representation (a firm price, a confirmed booking, an entitlement or refund) to a human, or to a script you've verified against your actual policies. Align the bot's answers with the terms you genuinely offer, log every conversation so you can see what it's telling people, and pressure-test it with the awkward questions customers really ask. Do that once, properly, and you keep the after-hours lead capture without signing up for the after-hours liability.

Key takeaways

Common questions

Related notes

Not sure what your AI is allowed to promise your customers?

A short conversation can map where your chatbot or phone agent makes representations on your behalf — and turn it into a clear line between what it answers and what it hands to a human. Book a call.