Tomb Raider's AI label is a lesson in saying how you use AI before someone else finds it

At the start of this month, Crystal Dynamics showed a new trailer for Tomb Raider: Legacy of Atlantis during Sony’s State of Play showcase. Fans mostly liked it. Within a day, people found an AI content disclosure on the game’s Steam page, and a lot of the conversation moved from the trailer to the label.

The disclosure said AI-assisted tools were used for early exploration and temporary development content, and that those assets were replaced or refined by humans. The studio then told Eurogamer it uses AI to help teams iterate faster, while ensuring “all finished content in the final product is human-crafted.”

That is a reasonable policy, stated clearly. It still turned a good trailer week into an AI story. I have spent my career where media meets technology, and right now I run VLO Studios, a team of about twenty people building games inside Fortnite. So I watch these moments closely. The lesson here applies far outside games.

The disclosure came after the reveal, and that was the problem

Look at the order of events. The audience saw the work first, liked it, and then found a label that seemed to change what they had just seen. Even though the label described early-stage use, the sequence made it feel like a reveal of something hidden.

Compare that to what happened with Crimson Desert in March. Players found what looked like AI-generated paintings in the game shortly after it launched. Pearl Abyss confirmed the art had been unintentionally included in the final release. Some 2D props had been made with experimental AI tools early in development to explore tone and atmosphere, and they were supposed to be replaced before launch. Nobody replaced them. The studio apologized, promised a full audit of in-game assets and said plainly that it should have disclosed its AI use.

TimelineHow AI disclosure reached game stores

Jan 2024Steam adds AI disclosure rules
Jan 2026Valve exempts coding tools
Mar 2026Crimson Desert AI art found
Jun 3Tomb Raider AI note reported
Jun 16Survey: AI wording a turnoff

Those are two different failures. Crimson Desert was a process failure: placeholder work that nobody tracked shipped to customers. Tomb Raider was a communication failure, a small one: the right policy, explained after the audience had already reacted. Both studios used AI the same way, as a sketchpad during development. One ended in a public apology and an audit of every asset. The other cost a news cycle.

What the platforms already require

Valve has required disclosure on Steam since January 2024. Developers fill in a section of a content survey that separates pre-generated content, made with AI during development, from live-generated content, made by AI while the game runs. Players see a note on the store page. Games that generate content live must also describe the guardrails that keep the AI from producing illegal material.

In January this year Valve clarified the rule: tools that never show up in the final product, such as coding assistants and office software, do not need disclosure. Artwork, sound, text and story that players actually experience do.

The live-generated category is about to matter a lot more for the kind of work my studio does. In April, Epic added an experimental feature to UEFN, the toolset for building inside Fortnite, that lets creators make AI-voiced characters players can talk to. For now creators can only test it, and islands using it cannot go live on public servers until the feature is more stable. When it does go live, every studio using it will face the harder version of this question. You cannot check every line an AI character says before a player hears it, so the disclosure and the guardrails have to be decided before launch, not after the first strange clip shows up online.

I think Valve’s line is the right one for almost any business, games or not. Nobody cares that your accountant used AI to reconcile a spreadsheet. People do care when the thing they paid for, read, watched or listened to was made by a machine and nobody told them.

Before and after

Used in the work

Code assistants
Drafts and sketches
Internal summaries
Research

Shipped to customers

Final art and video
Published copy
Voices and music
Product photos

The right column is where customers expect to be told.

Customers are paying attention to the word itself

On the same day this post goes up, TechCrunch reported a WordPress VIP survey of 2,000 people, including 1,200 US adults. Sixty percent of US consumers said brands using “AI” in their messaging are a turnoff, and 86% said they don’t fully trust AI and still want to check original sources.

That does not mean you should hide AI use. It means the opposite. If the word itself makes people wary, the worst version is the one where they discover it on their own. The best version is a plain sentence, in the place they would look, before they look: what you used AI for, what a person made or checked, and who to ask.

Both game studios now have that sentence. They just wrote it after the fact.

The same goes for anyone who makes work for clients. If you run an agency or a production company and a client learns from someone else that their campaign images were AI-made, the meeting that follows will be about trust. Put a line in the statement of work that says where AI is used and who checks the result, and the question never becomes a surprise.

The setup every company needs

You do not need a game studio for this to apply. Any company putting AI into marketing, product photos, support replies or published content has the same two exposures: something AI-made ships when it was not meant to, or something ships correctly and the explanation arrives late.

Here is the setup I would put in place:

Tag AI-made drafts and placeholders at the moment they are created, in the file name or the asset system, so nobody has to remember later.
Add one check before anything customer-facing ships: has every tagged placeholder been replaced or approved by a named person?
Write your disclosure in two or three plain sentences now, while nobody is asking, and decide where it lives.
Decide who answers when a customer, journalist or partner asks how you use AI, and give them the same sentences.

None of this slows down the work. Crimson Desert’s problem was a missing check, and Tomb Raider’s was timing. Both are fixed by things you decide once.

ChecklistBefore AI-touched work ships

Placeholders are tagged at creation
A person signed off on each one
Your disclosure is already written
Someone owns the questions

What to do this week

List every place AI touches something a customer sees: ads, product images, emails, help articles, video, voice.
For each one, write down whether a person reviews the final version, and who.
Add a placeholder tag to your asset naming or project tool, and tell the team to use it starting today.
Draft your disclosure in plain words and put it where customers would look for it, such as your FAQ or the product page.
Ask one person outside the team to read it and tell you what they think you meant.