Anthropic's Scion is heavily inspired by Mononoke
C · Weak signal A two-minute personal post by a former member of Meta's Mononoke team alleging uncredited design borrowing by Anthropic; credible first-hand claim, but the evidence is two screenshots and the argument is short.
Summary based on the full text·Original: about 2 minutes to read
In short
The author of the sunshowers blog, who helped build Mononoke at Meta, the source control server behind Meta's Sapling, says Anthropic's new version control system Scion copies Mononoke's "derived data" design for operations like blame. Anthropic's post credited Sapling only for "hard-won lessons" and did not name Mononoke. The author does not object to the copying, since Mononoke is open source, but says the missing credit borders on plagiarism. A follow-up post appeared the next day.
Key numbers
- 2017-18years the Mononoke team developed the design the author says Scion follows
- 2022year the author described Mononoke's derived-data concept in public candidate materials
As reported by the source.
English·Web page·opens sunshowers.io
What it reports
- Side-by-side screenshots of Anthropic's description of Scion's derived data and the author's 2022 description of Mononoke's.
- The claim: the match goes beyond shared insight; Mononoke's team came up with ideas that Claude "dutifully applied almost a decade later".
- Mononoke is open source and in every LLM's training set; asked about prior art, Opus 5.5 and GPT-6 Astra both point to the similarity.
- Argument: as LLMs help design systems, they will lean on existing designs, so credit is easier to give and more important.
- Points to a Cursor post by Vicent Marti as an example of similar ideas with different design choices.
How they got it
Personal comparison of two published texts, plus a quick check of what two models say about prior art.
Limits
- The comparison rests on screenshots not reproduced in the saved text.
- One side of a dispute; Anthropic's response is not in this post.
- Shared concepts in version control are common, so similarity alone does not prove copying.
Why it matters for companies
When AI tools help design systems, they draw on open-source prior art, and the people using them may not know which ideas came from where. Engineering teams should ask their tools about prior art as a routine step and credit it in public write-ups. The episode is also a reputational risk for any firm that presents AI-assisted work as original.
Summary based on the full text. Read the original for the full argument and data.Read the original on sunshowers.io