What Designers Should Know

Every time a designer uploads work to a portfolio site, publishes a Figma community file, or ships an asset to a stock library, there's a real chance that work is being used to train the next generation of AI design models, often without their explicit consent. AI models require vast amounts of data to learn how to generate designs, UI layouts, and illustrations, and that data often comes from the same publicly available images, creative communities, and software tools designers use every day.

How AI Uses Design Work

Adobe faced significant backlash after users discovered Firefly was trained on Adobe Stock and Behance content. Dribbble has not explicitly confirmed whether its content is used for AI training. While Figma has not disclosed its AI data use, its community files and third-party plugins likely contribute to training datasets, meaning the exact files designers share to help each other work faster may also be feeding the models built to compete with them.

The Legal Battles Have Begun

The most closely watched training-data case in design so far reached a verdict in November 2025: Getty Images v. Stability AI, decided in the UK's High Court. Getty alleged Stability AI copied more than 12 million of its photographs, along with metadata and descriptions, to train Stable Diffusion. The court rejected Getty's core copyright claim, because the model doesn't store training images directly, it couldn't be proven that outputs were directly derived from Getty's specific photos, but did find limited trademark infringement, since some early Stable Diffusion outputs reproduced watermarks resembling Getty's and iStock's. For designers, the practical takeaway is sobering: the ruling suggests training on copyrighted images may not itself be infringement under UK law, even though it can still create downstream trademark exposure.

The New York Times' lawsuit against OpenAI and Microsoft, filed in December 2023 over the use of copyrighted journalism in ChatGPT's training data, remains active and unresolved, still in discovery as of mid-2026, with no trial date set. As recently as July 2026, the Times and other publishers filed a legal memorandum accusing OpenAI of withholding evidence about how its systems were trained, requesting court sanctions. A separate class-action still accuses Stability AI, Midjourney, and DeviantArt of scraping millions of artworks without consent. Together, these cases make one thing clear: the legal framework for consent and compensation in AI training data is still being written in real time, case by case, and outcomes are genuinely inconsistent across jurisdictions.

If You Share Work Publicly

The Getty ruling in particular changes the calculus for designers who assumed copyright alone would eventually protect their published work from being used as training data. It may not, at least not automatically, and not everywhere. That doesn't mean there's nothing you can do; it means the protection has to be more deliberate than simply owning the copyright and hoping the law catches up.

How to Protect Your Designs

• Opt out of AI training when possible
• Make work private on portfolio sites
• Read the terms of service carefully
• Use watermarks or AI-blocking tools

None of these steps are foolproof, but together they meaningfully reduce how much of a designer's public output ends up in a training set without their knowledge. Watermarking in particular has an interesting secondary benefit post-Getty: the ruling's finding on trademark infringement suggests visible marks can matter legally in ways invisible copyright claims currently don't.

How Platforms Respond

Different platforms have taken different stances, and the differences matter for where designers choose to publish. Adobe now offers an explicit opt-out setting for Adobe Stock contributors and has stated that Firefly is trained only on licensed and public-domain content going forward, following the earlier backlash over Behance and Stock data. Other platforms have been far less transparent, which is itself informative, a platform's silence on its AI training policy is generally not a sign that one doesn't exist.

This unevenness has pushed some of the response outside individual platforms entirely. Artist and designer advocacy groups have pushed for industry-wide standards around consent and attribution, and tools like Glaze and Nightshade, built by researchers specifically to make images resistant to being cleanly used as AI training data, have seen real adoption among illustrators and designers who want a technical safeguard rather than relying solely on a platform's policy.

What Designers Can Expect

None of the current legal uncertainty is likely to resolve quickly. The Getty and New York Times cases alone could take years to reach final rulings, and even once they do, an outcome in one jurisdiction won't automatically apply in another. In the meantime, the most realistic posture for a working designer is treating AI training risk the way you'd treat any other business risk, not ignorable, not paralyzing, just something to factor into where and how you publish, the same way you'd think about licensing terms or client contracts.

Building AI Awareness

Beyond individual protective steps, the designers navigating this best treat AI-training risk as an ongoing part of their practice rather than a one-time settings change. That means periodically re-checking a platform's terms of service, since policies shift as litigation and public pressure evolve, Adobe's current opt-out options didn't exist a few years ago, and other platforms will likely follow a similar path under enough pressure. It also means being deliberate about what gets published where: portfolio pieces meant to attract clients may be worth the exposure, while process files, unused concepts, or client work under NDA generally shouldn't be public regardless of AI training concerns.

Frequently Asked Questions

Does opting out work?

It reduces the risk but doesn't eliminate it. Opt-out settings only apply to platforms that respect them, and don't retroactively remove work already scraped before the setting was enabled. Combining opt-out settings with private portfolios for your most sensitive work is a more reliable approach than either alone.

Is it legal for AI companies to train on publicly posted design work?

It depends heavily on jurisdiction and is actively being litigated. The Getty v. Stability AI ruling found no copyright infringement under UK law in that specific case, but the outcome could differ under US law, where the New York Times case is still unresolved, so "legal" currently varies by court and by case.

Should I stop posting work?

Not necessarily, visibility still matters for building a career, and pulling everything private has real professional costs. The more practical approach is being deliberate about which platforms you use, checking their AI training policies, and treating watermarking and licensing terms as part of your workflow rather than an afterthought.

Do watermarks actually work?

Imperfectly, but not uselessly. Tools like Glaze work by subtly perturbing an image in ways that are invisible to a human viewer but disrupt how an AI model interprets it during training, which raises the cost and difficulty of using that image cleanly. None of these tools offer a guarantee, and techniques on both sides continue to evolve, but combined with an opt-out setting and a platform's stated policy, they meaningfully shift the odds rather than leaving your work completely unprotected.

Does this cover client work?

Yes, and arguably more so. Client work typically carries contractual obligations around confidentiality and ownership that make unauthorized AI training use a bigger liability question, not just a personal one. Adding explicit language to client contracts about how deliverables can and can't be used for AI training closes that gap, since most standard contract templates predate this issue entirely and say nothing about it either way.

Is there a way to check whether my specific work has already been used for AI training?

Not reliably, in most cases. A handful of tools let you search whether an image resembles content in a known training dataset, but coverage is partial and most major AI labs don't publish searchable records of their full training sets. That gap is itself one of the central complaints in the ongoing lawsuits: without transparency into what a model was actually trained on, individual creators have very little ability to verify their own exposure, let alone seek redress for it.

AI and Design Ethics

The intersection of AI and design raises complex ethical and legal questions that won't be settled quickly. Designers must remain informed about how their work is being used, not because the answers are simple, but because the platforms hosting their work rarely make the trade-offs obvious. The same accountability gap shows up across other domains too, which is part of why we've argued that AI regulation needs to move faster than it currently is. If you're navigating how AI fits into your team's design process rather than just your personal portfolio, that's exactly the kind of question our product design work is built to help answer.