The article.
Published 10 October 2026
A promise your AI assistant makes to a client binds your business.
"The AI agent is the business."
Two Australian legal experts told SmartCompany on 9 October 2026 that the business, not the software, answers for an AI agent. Simon Newcomb, partner and head of AI at Clayton Utz, was one. Jeannie Paterson, professor of law at the University of Melbourne, said "there's no doubt at all that the responsibility lies with the person that's deployed the agent."
That view changes how a salon, a clinic, a studio or a consultancy should set up an AI assistant. An assistant answering enquiries on a website, in Instagram messages or in an inbox is not a helpful extra beside the business. Every answer the assistant gives a client is an answer from the business.
Why a business cannot blame the software
ChatGPT, Gemini, Copilot or Claude answering enquiries in your name has no legal standing apart from the business. Newcomb told SmartCompany there is "no legal difference between the agent and the business."
SmartCompany pointed to a case from Canada. An Air Canada chatbot told a customer he could claim a bereavement fare after he had travelled. British Columbia's Civil Resolution Tribunal rejected the airline's argument that the chatbot was responsible for the chatbot's own actions. Air Canada was ordered to compensate the customer.
Closer to home, the Australian Competition and Consumer Commission told SmartCompany that a business can be liable for misleading information through any channel, AI chatbots included. So an assistant should answer only what the business can stand behind.
Where a service business is exposed
A service business sells on promises. A booking time, a price, a result, a refund policy and a cancellation fee are all promises. An assistant can get any one of those promises wrong, late at night, in a perfectly friendly tone.
The answers that cause trouble are ordinary ones:
- An assistant quotes last year's price.
- An assistant confirms a slot your calendar already filled.
- An assistant offers a discount to close an enquiry.
- An assistant tells a client a treatment suits her skin.
- An assistant agrees to a refund your policy does not allow.
Newcomb flagged the risk of an assistant replying in real time. Nobody gets the chance to check the reply before the client relies on the answer.
Three limits to write down before an assistant answers a client
Newcomb's advice to businesses was to limit what an agent can discuss, control what information the agent can reach, limit the replies the agent can give, and keep testing, because the AI models underneath keep changing. For a service founder, that advice becomes three written limits.
Three written limits
01
What the assistant may answer
Opening hours, services and how to book sit with the assistant. Price exceptions, refunds, professional advice and complaints come to you, with a set sentence the assistant uses to hand them over.
02
One source for prices
The assistant works from the same price list your booking page shows, updated the day prices change, never a copy pasted into a prompt months ago.
03
A regular test
At least monthly, and again whenever the AI model changes. Ask the ten questions clients ask most, and read every answer as though a client already relied on the reply.
A page of written limits takes an afternoon. A client holding a screenshot of a promise your business never meant to make takes far longer to sort out.
Have a person approve drafts wherever a reply can wait
For posts, emails and quotes, a draft that waits for a person never reaches a client unchecked. EMME Influence content goes through the same approval step. A post reaches Instagram, Facebook or LinkedIn only after Melanie Laas approves the draft.
Live chat is where approval is hardest, so live chat is where the written limits matter most. Where a reply can wait ten minutes for your eyes, let the reply wait.
A booking assistant can still do most of the work inside those limits. The assistant offers the open times your calendar shows and the prices your booking page lists. Anything outside those two sources becomes a note to you, not an answer to the client. The client still hears back quickly, and you decide every exception yourself.
Check the contract with whoever built your assistant
Newcomb told SmartCompany that "contracting with a service provider to supply an agentic solution is outsourcing your governance." His advice was to settle the contract terms with that provider before the assistant goes live. The contract decides who pays when the assistant gets an answer wrong.
If a developer or an agency set up your assistant, ask three questions before the next client conversation.
Clients may also send their own AI agents to book and buy for them. Newcomb suggested updating your terms and conditions to say who is responsible for a purchase made by a customer's AI agent. Paterson said a business may be able to undo a deal where a customer knowingly exploited an agent's mistake.
An assistant with written limits gives a founder her evenings back. An assistant without written limits makes promises in her name while she sleeps.
This article is general information, not legal advice. A lawyer can apply these points to your own business and your own contracts.