Blog

Claude's rules on checking AI work and telling clients

By Prompt Works Studio, Toowoomba · Published 3 November 2026
Drafted with AI, approved by Prompt Works Studio.

Picture a Wednesday night. You run a busy café and the next fortnight's roster is due. You ask Claude to sort it out around everyone's availability, and it does a tidy job in a couple of minutes. You paste it into the staff group chat and go to bed. In the morning one of your casuals has lost two shifts, and nobody can tell them why.

That's a made-up Wednesday, but it's the kind of job some owners already hand to AI. It's also one that Anthropic, the company behind Claude, has rules about. Its updated Usage Policy takes effect on 12 November. Anthropic says the rules on checking and telling people haven't changed. What it's done is rewrite that part so it's clearer which jobs it covers. So now's a good time to read it.

My short answer is this. If Claude helps decide something about a person's legal rights, finances, health or job, have someone who knows the field check it before it goes out, and tell the person AI was used. Put that in writing for your team.

What does the rule actually say?

Anthropic calls these high-risk uses. They're the ones that can affect someone's health, legal rights, finances, livelihood, or access to housing, education or essential services.

When Claude is used for one of those, two things have to happen. First, a qualified person has to properly review what Claude came up with before it's given to anyone as advice or used to make a decision. That person needs the training or experience to judge it, and a licence where the law needs one. They can change it, and they stay responsible for it.

Second, the person on the receiving end has to be clearly told that AI was used. You don't have to say which AI.

Does it cover my business?

It might, even if you don't think of your work as high-risk. The policy lists the areas, and a few of them touch everyday small businesses:

• Legal: advising someone on their legal rights or what to do in their matter, or preparing legal documents for them to sign or file.
• Finance: advising someone on their own taxes, or preparing a tax return for them to sign or file.
• Employment: screening or rejecting job applicants, setting or changing someone's shifts, hours or pay, or judging how a worker's going.
• Housing: approving or knocking back a rental application.

So the accountant using Claude to prepare a client's tax return is covered. So is the café owner using it for the roster. If you're not sure which of your jobs need a person checking, that's what a Workflow Audit looks at, job by job.

What doesn't it cover?

Quite a lot of the useful stuff. The policy says the rules don't apply to internal drafting, research, summarising or analysis that isn't the final recommendation. They also don't apply to general information, like explaining what a law says without applying it to someone's situation, or to business jobs that don't advise or decide about a particular person.

So a law firm asking Claude to summarise a long file for the solicitor sits outside these rules, though the firm's privacy duties still apply to what goes in. A bookkeeper asking it to explain a new tax rule in plain words is outside them too. The rules kick in when what AI came up with goes to a person as advice, or becomes a decision about them.

There's one more rule worth knowing. If you put a chatbot built on Claude on your website to talk to customers, it has to tell them they're talking to AI. That goes at the start of each chat or somewhere on the chat screen.

Is this just Anthropic's rule?

Anthropic says these rules don't replace local laws, and where a local law asks for more, you follow the law.

Here, the Office of the Australian Information Commissioner (OAIC) wants businesses to explain clearly in their privacy policies and notices how they use AI. It also says public-facing AI tools like chatbots should be clearly marked as AI.

What should you check before AI work goes to a client?

Is the job on the policy's list? If it is, treat it as high-risk.

Who checks it? Name the person for each job, so it never falls between two people.

Is the check a proper one? Anthropic's word is "meaningfully". Someone reads it right through and fixes what's wrong before it goes.

How will you tell them? Work out where that goes, like the email, letter or report they get.

Will your team remember? An AI use policy puts the jobs that need a check, and who checks them, in writing.

Would you send a new starter's work out unchecked?

Think about a new staff member's first letter to a client. Someone senior reads it before it goes, because it goes out under the business's name. I'd treat AI's work the same way. It can do the first draft fast, and a person still puts their name to it.

Want to know where AI fits in your business? Book a call.