Flathub's New AI Policy Has Some Problems

Summary

The video analyzes Flathub’s updated policy on AI-generated and AI-assisted software. The speaker argues that the policy’s goal is understandable, but that some wording and exceptions could lead to inconsistent enforcement.

Policy background and scope

The speaker compares Flathub’s policy with policies for GNOME Extensions and GNOME Circle, which allow AI as a learning or development aid but reject submissions that appear substantially AI-generated or poorly understood by their authors.

The policy applies to software submitted to the Flathub store, not to Flatpak apps distributed through other stores. It covers both the app and the materials submitted to Flathub, including manifests, metadata, patches, build scripts, and pull requests.

Submissions must not be created or automated with AI tools or agents, and submitters should not ask AI tools to review their submission pull requests. AI-generated or AI-assisted code, documentation, and other content are also disallowed, as are contributions that are copyrighted, license-incompatible, or ethically questionable.

Violations may lead to rejected submissions or, after repeated violations, a ban. The policy is not applied retroactively to apps already published. It also allows exceptions for “mature, well-maintained projects.”

Why the policy was introduced

The video points to the volume of low-quality AI-generated submissions and the time volunteer reviewers spend evaluating them. Flathub contributor Bart said that LLM use is likely to continue growing, while also describing a rise in difficult interactions with submitters whose projects were rejected. The policy is presented as a response to this workload.

Main criticism: unclear boundaries and exceptions

The speaker considers the policy’s general aims reasonable, particularly its attention to low-quality code and license compatibility. However, he argues that terms such as “AI-assisted,” “mature,” and “well-maintained” are not defined clearly enough.

• It is unclear where ordinary autocomplete or code-completion tools fall on the spectrum of AI assistance.
• The mature-project exception lacks explicit criteria. The speaker mentions community engagement, release cadence, CI, development history, and project health as possible considerations, but says these should be documented if they guide decisions.
• Because the policy is not retroactive and allows exceptions, it could create a two-tier system in which existing or established projects continue using AI-assisted code while new submissions face stricter rules.
• Vague standards could lead to inconsistent decisions between reviewers and disputes over whether a project qualifies for an exception. Clearer criteria could reduce confusion and make rejection decisions easier to explain.

AI use across software dependencies

The speaker surveys popular Flathub apps and points to projects that disclose AI use, including Telegram Desktop, Lutris, and Gear Lever. He also argues that apps may indirectly include AI-assisted code through shared technologies and dependencies, citing browsers, Electron- and Chromium-based software, Qt-based applications, the Linux kernel, and systemd.

His broader point is that distinguishing a wholly AI-generated app from software that contains some AI-assisted code somewhere in its dependency stack may be difficult. He emphasizes that this is not an endorsement of “vibe coding,” but an observation about how widespread AI tools have become.

Overall assessment

The speaker concludes that Flathub’s policy is broadly understandable, but its vague language and undocumented exception criteria leave room for inconsistent enforcement and future disputes. He also notes that debates about the policy often become broader arguments about whether AI belongs in open-source development. Differing attitudes among future generations of developers may create further friction.

Reviews, guides, or tutorials

No product review, guide, or tutorial is presented. The video is a policy analysis.

Main speakers and sources

• Brodie Robertson — main speaker and commentator.
• Bart (Flathub contributor) — quoted from a Mastodon post about the policy and reviewer workload.
• Tom Hall / OSNews — cited for criticism that exceptions could create a “rules for thee” perception.