You will be able to apply a short set of rules that reduce copyright and ownership risk.
You've now seen that ownership of AI output is unsettled, in lesson 4.1, and that both inputs and outputs carry risks, in lesson 4.2, Risks on the way in and on the way out. With so much unsettled, you might reasonably wonder how you're supposed to know what's allowed at work on a Tuesday afternoon with a deadline?
You don't need the law settled to work sensibly. You need a few rules that keep you away from the riskiest situations and leave a trail if anyone ever asks questions. This lesson gives you four, and shows how Kelvin, the bakery's marketer from lesson 4.1, uses them.
Before anything else, check two documents: your organisation's policy on AI use, if it has one, and the terms of the tool you're using.
The employer's policy wins because it reflects decisions someone has already made about risk, often with legal advice you haven't seen. It might say AI images can be used on social media but not on packaging, or that client deliverables need a manager's sign-off if AI was involved. It protects you personally as well, because if something goes wrong while you were following the policy, the questions are about the policy and whoever wrote it.
The tool's terms matter for the reason in lesson 4.2: they set out what you may do with output. Some restrict commercial use on free plans. Some require you to say content was AI-generated in certain contexts.
If your workplace has no policy, that's worth raising with your manager. Module 8 helps you write one.
For anything commercial, don't ask for work "in the style of" a named living artist, illustrator, photographer or writer, and don't ask for anything that imitates a specific brand, logo, character or product design.
Style alone isn't generally protected by copyright in many countries. But prompting by name makes it much more likely the output will reproduce something distinctive from that person's actual work, and it invites a complaint even where the law might be on your side. Artists have been vocal about this, and the reputational cost of being the business that copied a local illustrator's look can be worse than any legal risk.
Describe what you want instead. Kelvin stopped writing "in the style of" followed by a well-known animation studio. He now writes "soft watercolour, warm evening light, hand-drawn feel, muted reds and golds". He gets images he likes, and they don't lean on anyone else's signature.
For anything important that's going out in public, run a quick check for close matches.
For images, use a reverse image search, such as Google Lens or TinEye. Upload the image and see whether anything very similar already exists. Kelvin does this for every image that goes on packaging or in paid ads. Once, it showed that his generated mascot looked strikingly like a character from a cartoon series, which he hadn't recognised, so he redrew the mascot's face and ears before anything was printed.
For text, a plagiarism checker can flag passages that match existing published writing. Many universities and businesses already have access to one. It's most useful for longer pieces such as articles, reports and product descriptions.
Neither check is perfect. A clean result means no close match was found, not that the work is guaranteed safe. But it catches the most embarrassing problems, and lesson 6.3, Checking images and video, goes deeper into image search.
For anything you publish or sell, keep a simple record of how it was made: the tool and plan, the prompts you used, the output you started from, and what you changed by hand.
This does two jobs. First, as lesson 4.1 explained, how much human creative input went into a work may matter to whether you hold rights in it. A record of your own edits is evidence of that input. Second, if someone raises a complaint, you can show exactly how the work was made, rather than trying to remember months later.
The record doesn't need to be elaborate. Kelvin keeps a shared folder for each campaign. In it goes a short text file with the prompts, the raw generated images, and his edited versions. It takes him about two minutes per piece.
The four rules work best as a checklist you run before anything AI-assisted leaves your hands. Kelvin's version for the mooncake campaign took less than ten minutes: confirm the bakery's own rule allows AI images in print, confirm the image tool's plan allows commercial use, check that no prompt named an artist or brand, run a reverse image search on the final design, and save the prompts and his edits to the campaign folder.
None of this makes the legal questions from lesson 4.1 go away. What it does is keep you clear of the obvious problems, and leave you able to show your working if a less obvious one turns up.
Think of the next piece of AI-assisted content you're planning to put out, whether a post, a flyer, a slide or a product description. That's the piece to run the four rules against first.
Turn the four rules into a checklist and apply it to one piece of AI-assisted content you plan to publish.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).