Qwen Code: Only 0.8% of "Google" Lines Are Actually from Google

A byte-by-byte audit of the Qwen Code repository reveals a significant discrepancy between the labeling "Copyright Google LLC" and the reality of the code. Out of 1,110 files displaying this header, only 58 exactly match Google blobs, representing 5.2%, and just 0.8% of the lines under the "Google" header actually come from Google. The method relies on the history of Google Gemini CLI, weighs the results by lines, and includes a reverse check.

Very few overlaps despite 1,110 "Google" headers

Out of the 1,110 Qwen Code files bearing the "Copyright Google LLC" header, only 58 are identical byte-for-byte to files present in the studied period on Google's side, which is 5.2%. When weighted by the number of lines, only 0.8% of the lines under these headers actually correspond to bytes sourced from Google. Furthermore, the files from the "mem0" connector are found nowhere upstream, even though they carry the "Google" label.

Verification through git blobs and reverse control

The audit began by organizing the headers of each TypeScript file, then examined whether the byte sequences of these files were present in the commit history of Google Gemini CLI, using git blobs. This approach aimed to limit errors related to changes that occurred after a fork, as well as modifications of names or locations. The classification of cases distinguished strictly identical copies at the byte level, those modified but retaining the same location, those moved or altered, and those never present upstream. The results were weighted by the number of lines, and a reverse check ensured that no Google code was hidden under Qwen headers.

Context and scope: headers treated as boilerplate

Qwen Code stopped using Google 10 months ago. The audit concludes that copyright headers primarily serve as persistent boilerplate, without guaranteeing the actual provenance of the code. Review tools and processes mandate the presence of this block, but do not ensure its update or verification against the actual author. It is therefore recommended to conduct byte-level checks on any derived code before relying on the attribution indicated by the headers.