Strata's update script failed on a Git error: there was no common ancestor. When someone investigated, the cause became clear. The entire commit history had been rewritten to remove "Co-Authored by Claude" lines from commit descriptions. This is not a minor operation: when a repository's history is rewritten, commit hashes change and existing clones break. The update script is immediate evidence of that.

The question is why. The simplest explanation is that someone wanted to present the project as exclusively human work, removing an explicit trace of an LLM's contribution. There seems to be no neutral technical reason. If a project wants to stop using a co-authorship trailer, it can do so from a certain point forward, leaving earlier history intact. Rewriting the entire past is a different choice: it changes what the repository claims to be.

The person who reported the case is blunt: it is hard to imagine a reason other than an intention to be dishonest about the project's origins. Claude's co-authorship does not alter the code, but it alters the narrative. For anyone evaluating an open source project, history is a source of evidence: it shows who worked on the code, when, and with which processes. If that history can be rewritten after the fact, auditing becomes more fragile. And when LLMs enter the development pipeline, the distinction between code written by a person, assisted by a model, or generated automatically is not merely academic. It concerns responsibility, code review, and compliance.

The issue becomes more concrete for organizations managing on-prem or self-hosted deployments. In those contexts, teams often need to prove software provenance for security and audit reasons. If the repository has erased AI-related metadata, part of that trail is no longer verifiable. Only the statements of whoever rewrote the history remain.

Strata's case signals a broader attitude: adopting LLMs in development but making them invisible. This is not only about reputation. If AI assists code writing but its presence is removed from documentation, those who inherit the project lose useful information about where to focus controls. Generative models can introduce vulnerable patterns or non-obvious dependencies; hiding their involvement does not eliminate risks, it makes them harder to trace. A project that rewrites its own past does not only ask for trust in its present: it asks users to accept a history corrected after the fact. For those who use the software, that is not a reassuring premise.