AI4.Games

The same question kept surfacing this week across very different stories: how honest are the people building and selling AI tools? A survey put a number on developer sentiment. A rendering technology rewrote character art without asking. A coding vendor got caught relabeling open-source work. And through all of it, studios kept shipping with AI anyway, some openly and some not.

88% Want Disclosure. Crimson Desert Got Caught Without It. Epic’s CEO Disagrees.

A GamesIndustry.biz survey of 826 game industry workers found that 88.4% believe generative AI use should be disclosed on storefronts. Nearly half disagreed with Valve’s current Steam policy, which requires disclosure for AI-generated player-facing content but not for AI used as a production tool.

Epic CEO Tim Sweeney publicly argued the opposite, saying mandatory AI disclosure “makes no sense” and arguing the technology will become as ubiquitous in development as game engines, making a similar disclosure requirement impractical. The same week, players found what appears to be undisclosed AI-generated art in Crimson Desert (Pearl Abyss), which sold two million copies on its first day with no AI disclosure on its Steam page.

You’ve got a survey, a CEO taking the opposite position, and a live case landing in the same week. Disclosure policy is no longer theoretical.

DLSS 5: Neural Rendering Draws Praise and Backlash in Equal Measure

The controversy with DLSS 5 isn’t really about upscaling. It’s about whether a rendering tool can start making creative decisions the artist didn’t request, and who owns the output when it does.

Nvidia revealed the technology at GTC 2026 as a work-in-progress neural rendering system that operates at the geometry level rather than simply upscaling pixels. Capcom, Ubisoft, and Warner Bros. are listed as launch partners, and Digital Foundry along with a veteran game artist praised the in-person demos. But online backlash erupted after demo footage showed altered character models in Resident Evil: Requiem, faces and proportions visibly changed from the original art. Your tech art team is going to have opinions about that one.

Jensen Huang stated developers retain direct control at the geometry level and can fine-tune the AI to match their creative vision. The technology is optional and won’t ship until autumn. For studios evaluating DLSS 5 integration, that’s now a creative control question as much as a performance one.

Cursor’s “Proprietary” Composer 2 Turns Out to Be Open-Source Kimi K2.5

Cursor released Composer 2 with a new Glass interface, marketing it as a proprietary model. Within hours, community analysis of API identifiers revealed it appears to be a fine-tuned version of Moonshot AI’s Kimi K2.5, which carries a modified MIT license requiring attribution for commercial products exceeding $20 million in monthly revenue. Cursor had not provided it.

We’ve seen vendor transparency become a theme this week, and this one is concrete: a procurement team evaluating Cursor now has a real question to ask. Separately, Cursor did ship a genuine improvement to its agent’s self-summarization, cutting errors by 50% and token usage by 80%, according to the company. The technical progress is real. The attribution issue is also real. If your team is evaluating Cursor, follow it on AI4.games to track what your peers are watching.

GitHub Copilot Ships Its First Long-Term Support Model

Procurement teams at studios running formal approval cycles have been asking for this for a while: a stable model version with a published deprecation timeline they can actually plan against. GitHub just delivered it.

GPT-5.3-Codex is now designated as Copilot’s first long-term support model with guaranteed support through February 2027. That was the enterprise headline in an otherwise dense week of Copilot updates: the coding agent starts 50% faster, semantic code search is live, session logs now trace directly to commits, and Squad multi-agent coordination lets multiple agents work across a repository simultaneously. No other major AI coding vendor has yet shipped a comparable formal LTS model with a published deprecation timeline. It won’t be the last.

AI Coding at Scale: Krafton and Nexon Go on Record

According to GamesIndustry.biz, Krafton and Nexon have both “committed heavily to AI integration” in their development pipelines, in what the outlet describes as the most comprehensive industry investigation of AI coding utility published this year. The feature examined where these tools actually help and where they fall short, with named studios putting their adoption strategies on the record.

Studio Leaders Speak: The Creativity vs. Efficiency Tension

GamesIndustry.biz ran a dedicated AI Week (March 16-20) that put multiple executives on record. FunPlus SVP Alexandre Amancio (formerly of Ubisoft) framed AI as a statistical execution tool that democratizes production but risks subordinating creativity to efficiency. Former NCSoft president Songyee Yoon, now heading VC firm Principal Venture Partners, said AI is inseparable from game development while advocating for artist protections. Dr Tommy Thompson offered a 20-year historical perspective, pointing out that studios like EA and Square Enix have used AI for QA automation for decades, well before generative AI entered the conversation.

InnoGames Puts Production Numbers on the Record

We covered the InnoGames Sunrise Village story last week. This week, on-the-record interviews with the engineering director and senior lead artist, reported by GamesIndustry.biz, added the operational detail. According to those interviews, the custom Unity/GPT-4o toolset (“AI Stage Designer”) runs six production steps with a human checkpoint at each stage. The game would have been cancelled without it; it was financially unviable at its previous team size. Former team members, including senior lead artist Lemuel Wuibout, moved to new project Cozy Coast and said they were happy about it. InnoGames’ stated strategy is to use AI productivity gains for more projects, not headcount reduction.

The counterpoint from the same outlet’s survey: 84.2% of respondents consider AI-generated content in games unacceptable. Both things are true simultaneously, which is about where the industry is right now.

Claude Code Patches a High-Severity Vulnerability

Anthropic patched a workspace trust bypass in Claude Code, identified by the security community as CVE-2026-33068 (CVSS 7.7), that allowed malicious repositories to skip trust dialogs. The fix shipped in v2.1.81. On the feature side, Claude Code Channels launched in research preview, enabling persistent CLI sessions controllable via Telegram or Discord. A community-built Unreal MCP tool also surfaced, letting Claude Code generate UE5 materials from natural language prompts.

Quick Hits

Nintendo’s Animal Crossing PC port draws AI criticism. According to TimeExtension, the Animal Crossing: New Horizons native PC port reportedly used Claude Code in the porting process, drawing criticism over quality and security concerns.

Scenario cited by InnoGames for production pipeline. InnoGames named Scenario alongside Google Gemini and Claude as key tools in its AI-first Sunrise Village pipeline. Scenario’s enterprise pivot, with node-based workflows connecting 250+ models, continues to pick up named production endorsements.

Take-Two CEO calls AI-made GTA “laughable.” According to Kotaku, Strauss Zelnick expressed skepticism about AI’s ability to handle AAA creative complexity at Take-Two’s scale.

Microsoft confirms AI priority under new Xbox CEO. Asha Sharma took over as Xbox CEO with AI as a strategic focus. Gaming Copilot is coming to current-gen Xbox consoles in 2026, with Matt Booty clarifying the assistant targets production support, not player-facing content.

Midjourney V8 Alpha launches. Faster generation and native 2K resolution, but premium HD mode costs more per job. Professional AI artists working on concept art are reportedly staying on V7 for now.

Ludo AI opens API and MCP integration. The game ideation platform launched programmatic access and Claude integration, letting studios connect Ludo’s concept generation into existing workflows.