At Svyazi, a creative agency, we have been producing AI-assisted content for clients since 2022. In that time, the tool set has changed completely. The legal exposure hasn’t. We keep running into the same problem: A client may expect full ownership of deliverables, but the AI platform’s terms of service tell a different story.
AI has moved into business workflows faster than most teams have had time to set policies around it. The tools are genuinely useful. But the terms attached to those tools carry implications that don’t get discussed in onboarding decks — and a few are worth understanding before they become a client conversation you weren’t prepared to have.
Most major AI platforms include some version of a reassuring line: The company doesn’t claim ownership of your prompts or the content you generate. That sentence is technically accurate. It’s also incomplete.
What the terms usually do grant the platform is a broad, perpetual, royalty-free license to use your inputs and outputs — for improving the service, training models and, sometimes, for marketing. The platform isn’t the owner. It just has the right to use the material indefinitely and for free.
For an agency or in-house team, this creates a specific problem. If a client’s contract includes a standard IP warranty — a promise that the deliverable is original and unencumbered — that warranty may not hold if the content was produced through a platform that retains usage rights. Whether that rises to an actual legal conflict depends on the specifics, and it’s worth reviewing with counsel before assuming either way.
Copyright adds another layer. U.S. law doesn’t protect content that’s fully AI-generated without meaningful human creative input — the Copyright Office has declined to register works where a human couldn’t demonstrate sufficient creative contribution beyond writing a prompt. If you’re delivering content to a client who expects to own it and build on it commercially, the amount of human editing and judgment that went into the final product matters.
The practical answer isn’t to avoid AI. It’s to treat the model’s output as a draft that you shape, not a finished product you hand off.
A lot of what goes into AI tools on a given workday is sensitive: a client’s messaging framework, a competitive analysis, internal meeting notes repurposed into a brief. Most of this flows into platforms without much thought about what happens to it next.
AI platforms vary significantly on data handling. Some use inputs for model training by default unless users opt out. Some share data with third-party infrastructure providers. And default privacy settings on newer consumer-facing tools often favor openness over restriction — meaning, if a team member doesn’t actively configure the settings, the defaults may not align with what a client would expect.
The organizational risk here isn’t hypothetical. In 2023, engineers at Samsung uploaded proprietary source code to ChatGPT while troubleshooting an internal problem. The data was retained by the platform before the company had policies in place to prevent it.
For teams working with client data, the exposure tends to be subtler: a campaign concept a client hasn’t announced, or a brief that contains commercially sensitive background. None of this is catastrophic on its own. But it adds up if there’s no clear internal rule about what can and can’t go into an external AI tool.
One practical workaround: Where real client data isn’t necessary, use placeholder information — fictional names, dummy addresses, generic product descriptions. If the data isn’t real, a leak doesn’t cost much.
Nearly every major AI platform operates on an “as-is” basis. This means no warranties about accuracy, originality, fitness for purpose or non-infringement. If the output contains a factual error that ends up in a client deliverable, or if it closely reproduces something copyrighted, the platform bears no responsibility. The user does.
The outputs can be confidently wrong. They can reproduce patterns from training data in ways that create IP exposure for whoever publishes the result. And, unlike a contractor or freelancer, the platform won’t be named in any resulting dispute.
The legal principle is fairly consistent across jurisdictions: A decision made with the help of an algorithm is still a human decision. The person or organization that used the tool is responsible for the outcome. AI platforms are structured to reflect exactly that.
Think of AI output the way you’d think about work from a talented but overconfident junior staffer — good at generating material quickly, prone to presenting uncertain things with unearned confidence, and always in need of a final check before anything goes out the door. One useful technique: Run a claim or summary through a second model to catch errors or outdated information the first one missed. It’s not a guarantee, but it puts another layer of judgment between the draft and the client.
None of this requires a legal team or a new policy manual to start addressing. A few habits go a long way:
- Before adopting a new platform, check four things: what it says about content rights, how it handles user data, what the default privacy settings are, and what its liability caps look like. The terms are public. They take less time to skim than most vendor pitch decks.
- Compare the platform’s terms against your client contracts. If you’re giving IP warranties in your SOW, make sure the platforms you’re using don’t quietly undermine them.
- Set privacy settings on day one, not after a problem surfaces. This is a five-minute task that most teams skip.
- Keep real client data out of external AI tools unless you’ve verified the platform’s data handling meets the client’s standards. Where possible, work with anonymized or placeholder content.
- Don’t send AI output to a client without human review. This means more than a spell check — it means someone with subject matter knowledge has read it, verified the claims that matter, and takes responsibility for what’s in it.
AI is a fixture in this work now. That’s not going away, and most of the tools are genuinely useful. But the platforms that provide them were built to serve their own interests first — and their terms reflect that. The market is consolidated enough that individual users have limited leverage to negotiate different conditions.
What teams can control is how they work within those conditions. The platforms aren’t going to rewrite their terms on your behalf. But knowing what you’ve agreed to and making sure it aligns with what you’ve promised clients — that part is squarely yours to manage.
Ilya Zmienko is the founder of Svyazi, a creative agency working at the intersection of graphic design, communications and AI-assisted content. With 10 years of experience in creative marketing and digital communications, Zmienko has produced TEDx events and projects that have been recognized at creative festivals.
