Epic Director Challenges Valve on Steam AI Clone Protections

Epic Games online services product director Alexander Rehm has argued that Valve’s emerging answer to AI-assisted clone games on Steam does not adequately protect the smallest developers. His central objection is less about whether a studio should make a distinctive, high-quality game than whether a tiny team can afford the perpetual development needed to stay ahead after that game is copied.

The debate follows a change to Steam’s review process, made amid a growing flow of games that appear to be at least partly enabled by generative AI. Valve has said it does not intend to prescribe which development tools creators may use, while Steam game pages include AI-related disclosures. The review adjustment is intended to make processing more efficient while adding requirements, but it is not presented as a total ban on low-effort or AI-assisted releases.

That distinction is where the practical argument begins. Disclosing the use of AI can give a customer more information. It does not, on its own, determine whether a release is a close imitation, whether it is misleadingly presented, or whether a small original creator can withstand a swarm of similar listings.

A good game is not automatically a durable defense

Valve representatives have said they are still working on the issue and have encouraged developers to make original, high-quality games that are hard to clone. Rehm called that a reasonable approach for teams that have the time, money and staffing to keep building. But he said it is of limited help to a one- or two-person studio whose achievement may be a single clever mechanic.

That is an important separation of two ideas often rolled into one. Originality is about whether a game offers a distinctive concept, presentation or combination of systems. A moat, in business terms, is the advantage that makes it difficult for rivals to capture the same audience. A game can be genuinely inventive without having a large enough team to create years of updates, marketing, live support and follow-up features around it.

Rehm’s formulation is that “keep improving forever” can become a cost borne mainly by well-funded teams rather than a workable safeguard for independent creators. In other words, continual improvement may help preserve a game’s appeal, but it can also shift the burden of platform safety onto the developer most vulnerable to imitation.

That concern is particularly relevant to a storefront built around discovery. A creator who releases a novel idea must compete not only with other original games but potentially with titles using near-identical descriptions, themes, visual language or mechanics. Even if customers eventually distinguish the original from an imitation, attention around a launch window can be finite. The problem described here is therefore not merely artistic offense at copying; it is a question of visibility, timing and the resources required to respond.

For PC players, the stakes include the ease of finding the game they intended to buy. Steam is still a major route to independent releases, as illustrated by the arrival of smaller projects and demos such as Ghost Haunting’s PC demo. A crowded store can contain ambitious projects, experiments and unfinished work at once. The harder it is to distinguish those categories, the more discovery depends on a buyer’s research rather than clear storefront signals.

Disclosure, review and enforcement are different tools

The conversation can be clearer if these are treated as separate problems rather than one giant “AI problem.”

Disclosure tells customers that AI was used in some part of a game or its creation. It is a transparency measure.
Store review is a platform’s process for deciding whether a submission meets its requirements and can be listed. A slower or more detailed review can catch some concerns, but it requires standards and capacity.
Clone enforcement concerns what happens when a game is so similar to another work that it is allegedly deceptive, infringing, or otherwise unacceptable under platform rules.
Legal enforcement involves claims that may fall under copyright, fraud, extortion or other laws and can require evidence, legal process and the appropriate authorities.

These mechanisms overlap, but none replaces the others. A disclosure label does not stop a clone. Review without a clear rule may not resolve borderline cases. Legal remedies can be essential, but they may be slow or expensive for a small studio confronting a fast-moving launch problem.

Valve staff were also asked about reports that developers had been threatened with AI-made clones unless they paid money. Valve said it had not seen such reports and characterized extortion as a matter for law enforcement rather than the storefront. Rehm took issue with that framing, saying it made the platform appear to treat the matter as somebody else’s responsibility.

There is a genuine boundary question here. A store cannot substitute for police or courts, especially when facing an allegation of extortion. Yet a platform can still decide how it handles listings, impersonation, duplicate-looking products, suspicious submission patterns and reports from affected developers. Rehm’s criticism is fundamentally a demand for a more active platform role before or alongside a legal case, not an argument that a storefront should decide criminal guilt.

Epic’s position invites a difficult consistency test

The criticism of Valve has produced an immediate counterargument: Epic’s own policies and public messaging on generative AI are not a neat match for the stance Rehm is taking here.

Former Valve developer Chet Faliszek challenged Rehm’s characterization of Valve’s comments and pointed to Epic’s broader AI-first direction, an AI-mandatory game jam and a “vibe coded” launcher. Those descriptions were raised as criticisms in the exchange, rather than as a settled accounting of Epic’s strategy, but they highlight why the dispute has become about more than Steam review queues.

Epic CEO Tim Sweeney has repeatedly objected to Steam’s AI disclosures. The Epic Games Store does not mirror those disclosures, although Epic’s Fab marketplace requires AI disclosure. Sweeney has drawn a distinction between the two: Fab is a business-to-business asset marketplace, he said, where buyers may be operating under contractual obligations to understand the provenance of material used in their games.

Provenance means the documented origin and history of an asset: where it came from, how it was made and what rights or conditions attach to it. For a developer buying art, models or other production material, provenance can matter because the buyer may need confidence that it can legally use the asset in a commercial game. That is a different, narrower purpose from informing a consumer that a finished game used generative AI somewhere in its pipeline.

The distinction does not eliminate the apparent tension. If disclosure is necessary when professional buyers need to trace the origins of assets, critics can reasonably ask why customers choosing a finished game should receive less information. Conversely, requiring a label for every possible AI-assisted element does not automatically answer whether the game is exploitative, deceptive or a clone. The policy debate is about what each label is meant to accomplish and who needs the information.

Unreal Engine, user creation and the scale problem

Sweeney has also emphasized generative-AI capabilities in Unreal Engine 6. The supplied discussion notes that a video-game lawyer has warned that ignoring copycat potential could invite legal difficulty. That concern should be treated as analysis rather than a determination that a particular feature or product is unlawful. No specific lawsuit or outcome is established here.

Sweeney has separately discussed the uncertain legal backdrop around AI training and output. He pointed to a court ruling involving Anthropic in which liability was found over torrenting pirated material while training itself was treated as fair use, and he noted that other decisions have turned on the similarity between AI-generated output and copyrighted works. The takeaway is not that every use of AI is permitted or prohibited. It is that different conduct can raise different questions: the source material used for training, the acquisition of that material, and the resemblance of a resulting output can each be assessed separately.

That uncertainty matters for platforms as they add creation tools and open ecosystems. Epic’s broader effort to connect Unreal Engine tools with Fortnite has prompted questions about how the company would avoid a similar flood of low-effort or duplicative creations. Maximilian Ygdell, chairman of Chalmers Robotics Society, put the challenge bluntly in his response to Rehm: if AI trained on others’ work was previously acceptable to Epic, why is its ability to empower widespread copying now a concern?

It is a pointed question, but it should not obscure the legitimate issue Rehm raised. Tools that reduce the cost and skill required to produce a game can help creators build new work. The same reduction in cost can make imitation easier and increase the volume that storefront moderators must assess. A platform’s challenge is to preserve the former benefit without allowing the latter to overwhelm discovery and trust.

What a material answer would need to address

Valve has not claimed to have finished solving the problem, and its review adjustment is evidence that the existing process is still being refined. Rehm’s argument is that a solution cannot stop at asking creators to outwork copycats. The relevant policy questions are more concrete:

How quickly can a developer flag a title that appears confusingly close to its own upcoming or released game?
What evidence does a storefront require before it investigates a report or takes a listing down?
Can a platform recognize coordinated duplicate submissions or misleading attempts to exploit a known game’s name and audience?
Will customers receive useful information about AI use, and what exactly will such disclosures cover?
How can a small team receive help without having to finance an endless stream of additions merely to stay recognizable?

Those questions also apply to Epic and any company building AI-assisted development tools, creator ecosystems or major digital marketplaces. It is easy to state that developers should make original games; it is much harder to define fair, consistent intervention when the alleged copy is not identical enough for an obvious legal answer but is similar enough to confuse customers or drain attention from the original.

For now, the argument is notable because both sides expose a real weakness. Valve’s process is still a work in progress, and reassurance that quality will win out may not satisfy a solo creator facing rapid imitation. Epic’s criticism, meanwhile, will continue to face skepticism while its store does not match Steam’s consumer-facing disclosure approach and the company promotes generative-AI tools of its own. The broad issue is not whether technology can help make games; it is whether the platforms distributing and enabling those games will take meaningful responsibility when the same technology makes copying cheaper at scale.