Anthropic accused of passing off Meta code as Claude design
An ex-Meta engineer says Anthropic new version control system is a direct copy of Mononoke, raising questions about AI plagiarism.
Drama · Source: Hacker News
What happened
Anthropic recently unveiled Scion. They describe it as a brand new version control system. The company claims their AI model Claude designed the architecture. But an ex-Meta engineer named Rain says the design is far from original. Rain points out that Scion relies on a specific derived data concept to make operations like blames more efficient. This is the exact same architecture Rain and their team built for Meta Mononoke source control server back in 2017.
The core issue here is not the act of copying. Mononoke is an open source project. The creators released it specifically to benefit the broader software community through the spread of good ideas. The real problem is the complete lack of attribution. Anthropic credited a different system called Sapling for hard-won lessons. They failed to mention Mononoke at all. Rain states this behavior borders on outright plagiarism.
Anthropic is trying to frame Scion as a massive triumph of artificial intelligence system design. The reality is much less glamorous. Claude likely just regurgitated Mononoke architecture directly from its massive training data. Rain notes that finding the source material is incredibly easy. If you ask Opus 5.5 or GPT-6 Astra about prior art for this design, both models immediately flag the striking similarities to Mononoke. There is simply no excuse for failing to credit the original human engineers.
Key facts
- Scion — Anthropic new version control system
- Mononoke — Meta open source version control server from 2017
- Derived data — The core architectural concept both systems use for efficiency
- Opus 5.5 and GPT-6 Astra — Models that easily identify the architectural similarities
Why it matters
We are rapidly entering an era where AI models design complex backend systems for us. But we must remember that models do not invent these architectures from scratch. They index existing open source designs, strip away the context, and repackage them as novel solutions. If you use an LLM to architect your next major product feature, you take on the liability. You are legally and ethically responsible for figuring out exactly where that design came from. You cannot just blindly trust the model when it claims originality.
This dynamic fundamentally threatens the open source social contract. Developers open source their hard work to push the industry forward. In return, they expect basic attribution and respect. If AI companies continue using models to strip attribution from open source projects, top tier developers will simply stop sharing their code. We need automated attribution pipelines built directly into the generation process. Until then, the burden falls on founders to verify their AI generated code.
For builders
Audit AI architectures for prior art
Do not assume your AI generated a novel system. You must ask the model directly what prior art it used to generate the design. You pay the massive reputational cost if you launch a stolen architecture and get called out by the community.
Protect your open source projects
The current open source licenses rely heavily on human goodwill for attribution. Founders building developer tools need to think defensively about how AI models ingest their work. You lose your competitive moat if an AI company repackages your open source architecture without credit.
Leverage AI to find attribution
You can use the exact same tools to prevent plagiarism. Feed your AI generated architecture back into Opus 5.5 or GPT-6 Astra. Ask the models to identify the closest existing open source projects before you publish your technical blog posts.
My take
I build AI products every day, and I know these models are just stochastic parrots with incredible PR teams. Anthropic trying to pass off a decade-old Meta architecture as Claude genius is exactly why founders need to stay grounded. Give credit where it is due. If you try to erase the human engineers who actually built the foundation, the developer community will tear you apart.
Original reporting: Hacker News. This is my rewrite and opinion.