ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
Google Artemis adds source attribution: AI open-source reuse cannot be judged solely by licenses

Google Artemis adds source attribution: AI open-source reuse cannot be judged solely by licenses

AI information Admin 2 views

On September 11, 2026, mobile automation company Minitap published an article stating that Google's mobile device agent project Artemis used some code and proxy prompts from its open-source project mobile-use, but did not cite the source in the README at the time. On September 12, Artemis added a new commit to the main branch, adding signatures for Minitap and mobile-use to the README and related files. This change shows that since the code is open to Apache 2.0, it does not mean that a reuser can complete source governance with just one license.

What can public records confirm?

The repository records displayed by Minitap include multiple sets of comparable content: implementations connecting Android devices, Hopper proxy instructions, sample tasks, and common defects found in older versions. It also points out that Artemis's earlier pyproject.toml listed three mobile-use authors, but these names were later removed during a historical replacement. The public issue created on September 11 required projects to specify derivative relationships, restore applicable copyright and attribution information, and record the original authors and contributors.

These materials are sufficient to prove the existence of a source relationship between the two codes, but whether it constitutes legal infringement still depends on the specific documents, the original submission date, license obligations, and jurisdictional judgment. Minitap itself acknowledges that Artemis involved independent engineering work and states that the ranking results are self-reported by the project team and have not been independently verified. Therefore, the article cannot directly state one party's claim as a fact.

Adding signatures to 23 documents does not mean the dispute is completely over

In the submission on September 12 at 02:59 UTC, the Artemis repository modified 23 files: the README explicitly states that the project contains source code developed by Minitap, and several implementation files add notes and copyright information stating "partially derived from mobile-use." This move aligns with Minitap's core mission and allows later Artemis developers to trace upstream.

However, as of the time of verification, the relevant Issues were still open, and the repository had not seen any formal explanation for the change in the author list. Supplementary attribution solved the visibility issue of the current version, did not automatically answer whether previous versions fully complied with Apache 2.0 terms, nor could it replace mutual confirmation of the specific code scope.

AI projects must reuse open-source code with four checks

  1. Record source: Whether forking, copying directories, or importing fragments with a coding agent, keep upstream repositories, commit hashes, and modify records.
  2. Document-by-file verification obligation: Check LICENSE, NOTICE, copyright header, and change notes; do not place only one license with the same name in the repository root directory.
  3. Treat proxy output as regular code review: AI-generated or reposted code also needs similarity, dependency source, and license scanning; responsibility does not disappear just because "done by the model."
  4. Separate evaluation and attribution: Performance rankings require reproducible experiments and independent verification, while code sources rely on version history and authorization records; the two cannot replace each other.

Artemis's subsequent revisions are a positive signal and raise a real issue for the open-source AI team: models and proxies have increased reuse speed, but have not automatically filled in source explanations. Integrating traceability, attribution, and license checks into the merging process costs far less than tracing history after project release.

Recommended Tools

More