Steam's AI disclosures are now normal. That's exactly the problem

What happens to a warning label when nearly half of all products carry it? It stops warning. It becomes part of the packaging, something your eye slides past on the way to the Buy button. That, in our view, is the real story behind the latest number to come out of Steam.

What the data says

German outlet Golem analyzed 2,803 new Steam releases and found that 41.67% of them disclose the use of AI. The analysis also looks at what developers use the technology for, but we only have the headline figure, so we won't guess at the breakdown.

Separately, Austria's Der Standard argues that Steam has an AI-clone problem and that Valve wants to do (almost) nothing about it. Its point: when a game goes viral, copies follow quickly, and in the era of tools like Claude Code a clone can be built in hours. Solo developers and indie studios, it says, are increasingly the ones who suffer.

Disclosure being normal is not the same as disclosure being useful

Let's be clear about what 41.67% does and doesn't tell us. It suggests that declaring AI is routine among new releases. It does not tell us that 41.67% of games are "slop." Golem's own headline uses that word as a question, and the analysis is framed around what AI is used for. That distinction is everything. A studio using a model for placeholder text or code assistance is not the same as a store page assembled from generated art, generated copy and a generated trailer.

If a disclosure reads the same for a developer who used an assistant to speed up tooling and for someone mass-producing storefront filler, players learn nothing. We think that is the weakness of any simple yes-or-no disclosure: it measures honesty, not quality.

There's also an uncomfortable upside for bad actors. A common disclosure makes AI use feel unremarkable, and unremarkable is exactly what a flood of low-effort releases needs. The more normal the tag, the less stigma attaches to it, including for the cases where stigma was the point.

The clone problem is where disclosure fails hardest

Here is the part we find most telling. Der Standard's complaint isn't about whether a game admits to using AI. It's about copies of successful games appearing fast, and, honestly or not, an AI disclosure doesn't solve that. A clone can disclose everything and still be a clone.

Disclosure addresses a transparency problem. Cloning is an enforcement problem. Mixing the two lets platforms look responsive while the damage lands on the people least able to absorb it: small teams whose one hit gets duplicated before they can enjoy it. If Valve's answer really is close to "almost nothing," as the Austrian outlet's headline puts it, the disclosure data becomes a nice dashboard on top of an unaddressed issue.

What we'd want instead

Our verdict: disclosure was a reasonable first step, and the Golem numbers suggest developers are taking it seriously. But when a rule is met by roughly four in ten new releases, it is no longer much of a filter. Three things would make it worth more, and all of them are our opinions, not announcements:

• Granularity: say what AI touched (art, code, voice, text), not just that it did.
• Visibility: make it easy for players to act on the information.
• Enforcement against copycats, handled separately from disclosure.

Until then, our advice to players is simple: read the disclosure as the start of the question, not the answer. Check what the AI was used for, look at who made the game and how long they've been around, and treat a sudden lookalike of a viral hit with suspicion. The number is a useful snapshot of where the industry is. It just isn't a guarantee of quality.