AI now writes the code, the tests, the marketing scripts and the emails, and the person who gave it the prompt is left wondering whose work it actually is. Anything you automate has an authority sitting next to it, and with AI doing the building, who that authority is has stopped being obvious. I’m going to start with QA because that’s the work I do, but the same question applies to anything you hand to a chatbot. Replace QA with writing, coding or a sales email and the answer barely changes.
Five Words People Mix Up When They Talk About AI Authorship
Most of the confusion in this topic comes from using one word for five different things. I’m not a lawyer and none of this is legal advice. I’m just keeping the terms apart so the rest of this post means one thing at a time.
- Author: who made the work and gets the credit.
- Copyright: the legal protection on a work. In the US it needs a human author, so purely AI output may get little or none.
- Ownership: who controls the work and can use it, which comes from employment, contracts and the terms of the tool you used. It isn’t the same thing as copyright.
- License: the permission attached to code or content you built from, open source or proprietary.
- Accountability: who answers when it breaks or causes harm. It’s separate from the other four, and it’s the one that follows you.
Credit What You Built From, Own What You Added
If you build a test case yourself, you can claim ownership of it. Say the team gives you a template, or the template is company property, and you add your own modifications. You credit the template and reuse it, and the modifications are yours now. The template is the inspiration and you acknowledge that you based your work on it, and that’s how it should be.
Code works the same way. You credit the app that inspired it, you credit the language you built it in, and it’s still yours when the building was yours. Credit goes to what you built from, and the work belongs to the person who did the building and decided what it should do.
When the AI Did the Building
Now let AI do the building. Say it’s vibe coded and the AI wrote the code, so who gets the ownership? My answer is the same one I’ve been giving on EngineeredAI: the AI is a tool, so it gets credited the way the template and the language do, and the human who directed it and checked it owns the outcome. I made that case in why AI is a tool and not a threat, where the human still sets the vision, applies judgment and owns the result.
Author, owner and accountable person can be three different answers here, and that’s the new part. You can use AI to build something, credit it the way you’d credit the language, and still be the one who answers for what it does. That comes from directing it, scoping it and verifying it, not from typing it.
The coding tools have started putting this in writing. Claude Code adds a Co-Authored-By: Claude line to the commits it makes by default, and Codex CLI has a similar setting that adds a Co-authored-by: Codex line, with a config to rename it or turn it off. GitHub shows those co-authors on the commit, so the tool is crediting itself the way you’d credit a template or a language. That’s credit and nothing more, and it doesn’t change that your name is on the commit and you’re the one answering for it.
The Same Question Shows Up in Everything You Hand to AI
Take QA, because it’s the version I live in. Testing used to be manual, then automation, and now AI writes a lot of the automation. You open Claude or ChatGPT, give it the steps, and it writes the Playwright or Cypress test in JavaScript, TypeScript or Python. You did the prompting and you supplied the steps, but the tool wrote the code, so the question of what you own is real.
What you own there is the part the AI can’t see. You own what correct means, because the tests it writes agree with the spec it was given and the spec isn’t the whole truth, which is the problem in AI testing passes the spec and still misses what you know is real. You own where it’s allowed to change things, and you own the call to accept the result. A test that passes only tells you the code agrees with the test, and both came from the same system.
Swap the work and nothing else changes. A marketing script, an email, a blog post or a function all come back from the chatbot looking finished. Whoever sent it, signed it or shipped it is the one holding the result, because the AI can’t answer for it.
What This Looks Like on a Real Project
On my QA retainer, the product is a SaaS built and maintained by an AI dev agent, and I’m the human verification layer. Every ticket covers one surface and carries a scope boundary, and every fix gets a verification comment before it closes. The AI tested what it built, and the comment is how I put my name on the result. The full loop is in the hybrid QA workflow for AI generated code.
It’s also why code from an AI dev team can’t be treated like hand written code. The code reflects a model’s reading of the prompt and not everything the team discussed, which I covered in happy path testing when your dev team uses AI. That gap is where the human still earns the ownership.
Can You Copyright AI Generated Content?
Here’s where it stops being opinion. The US Copyright Office looked at this in Part 2 of its AI and copyright report, published in January 2025, and the short version is that copyright protects human authorship. A prompt on its own doesn’t count, because the Office treats prompts as instructions and not as the creative expression in the output. What can count is the human part, like original expression that shows up in the result, or your own modifications and arrangement of what the AI generated.
The same principle came out of Thaler v. Perlmutter, where a federal district court said human authorship is a bedrock requirement of copyright. That’s the US position and the law is still developing, so it can look different in other countries and it can change. For code and tests, it means output you took as is may have little or no copyright protection, which limits your ability to stop someone else copying it. It doesn’t mean you can’t use it, and a copyright question and a can I use this question are different ones, which is the part that confused me while I was researching this.
Who Owns AI Generated Content? What the Terms and Licenses Say
On the contract side, the big AI tools assign the output to you. Anthropic’s commercial terms say that, as between the parties and to the extent permitted by law, the customer owns the outputs. OpenAI’s terms say the same for ChatGPT, that you own the output and they assign you whatever rights they have in it, if any. Both carry the same hedge, which tells you a contract can only hand over rights that exist. The consumer terms for Claude.ai use the same “if any” wording.
Those same terms also put the checking on you. Anthropic’s terms say it’s the customer’s responsibility to evaluate whether outputs suit their use, including where human review is appropriate, and OpenAI’s say to evaluate output for accuracy and appropriateness, using human review as appropriate. OpenAI also says you’re responsible for the content, including making sure it doesn’t violate any applicable law. So the tool assigns you the output and the review responsibility together.
Both vendors also offer a form of legal cover to paying business customers. OpenAI’s Copyright Shield says it will defend covered customers and pay the costs if they face a copyright claim over output, but it applies to ChatGPT Enterprise and the API, not the free or Plus tiers. Anthropic’s commercial terms offer a similar defense for authorized use, with exclusions, and as far as I could find the consumer terms don’t include an equivalent. That’s a contract between you and the vendor about legal claims, and it doesn’t change who the author is or who answers for the work.
If you made it at work, the employer usually comes into this too. Under the US work for hire rule, what an employee creates within the scope of the job is generally the employer’s, whether it was typed by hand or generated. But that only moves rights that exist, so if there’s no copyright in pure AI output, there’s nothing for the company to take over.
Then there’s licensing. These models were trained on code that includes open source, and GitHub has acknowledged that Copilot can sometimes recite public code it was trained on. If that code is under a copyleft license like the GPL, the license obligations can follow it, and as far as I could find, courts haven’t settled how much of that applies to AI output. So generated code isn’t automatically clean just because it came out of a chat window, and checking what you ship is still your job.
One thing I didn’t research is whether responsibility for something illegal built with AI can move from the person who prompted it to their company. That’s a separate question from authorship, ownership and copyright, and I’m not going to guess at it. If it matters for your situation, it’s one for a lawyer.
Design and Writing Tools Say the Same Thing
Design tools follow the same pattern. Adobe says that, as between Adobe and the customer, the customer owns Firefly output and is responsible for how it’s used, and Firefly attaches Content Credentials so the file shows AI was involved, which is in Adobe’s Firefly legal FAQ. Canva’s AI Product Terms say you own your output unless it modifies or includes Canva library content, and Canva makes no copyright claim over it. Midjourney’s terms give paid subscribers ownership of their assets, and in the versions I found that’s ownership to the fullest extent the law allows, while free users get a noncommercial license instead. Canva’s own help page adds that many jurisdictions, including the US, don’t give copyright protection to AI created works, and that applies to generative AI as a whole.
Writing tools say it too. Notion AI’s terms say input and output are your customer data, that you’re responsible for making sure the output doesn’t break the law or infringe anyone’s rights, and that output may not be unique. Grammarly’s terms treat generative AI inputs and outputs as your user content, and Jasper’s terms treat the output as the customer’s property while not promising it’s original or free of infringement.
One detail I didn’t expect is that Grammarly’s and Notion’s terms, and Canva’s Magic Studio terms in the version I found, each say you can’t use the tool to mislead anyone that the output was solely human generated. So the vendors hand you the output and still care about who gets named as the author. That fits the point that credit, ownership and accountability are different things.
So What Do You Actually Own?
My answer is the outcome. The AI, the template and the language all get credit for what they contributed, and the person who directed the work, checked it and decided to ship it is the one holding the result. That holds whether the work is a test, a function, a script or an email.
I’m not saying AI output is yours in every legal sense, because it isn’t that simple and it depends on where you are, who you work for and which tool you used. What I’m saying is that nobody else is set up to answer for it. If your team has AI writing the work and nobody is sure who signs off, start with what an AI human dev team workflow looks like and go from there.
