Shuji Sado (佐渡 秀治)

The Legal Hack: Why U.S. Law Sees Open Source as “Permission,” Not a Contract In Japan, the common view is to treat an Open Source license as a license agreement, or a contract. This is also the case in the EU. However, in the United States—the origin point for almost every …

Ever wonder why US law treats Open Source licenses as "permission," but in the EU/Japan they're seen as contracts?
The answer is a fascinating Legal Hack in US Copyright Act that hackers used to build the entire Free movement. I wrote a deep dive on this secret history:
shujisado.org/2025/07/24/w...

How Can Open Source Projects Accept AI-Generated Code? — Lessons from QEMU’s Ban Policy QEMU has formally adopted a policy that rejects contributions containing code generated by AI tools. The core reason is the concern that such AI-generated code cannot satisfy the requirements of th…

QEMU has formally adopted a policy that rejects contributions containing code generated by AI tools. The main reason is that such AI-generated code might not satisfy the requirements of the DCO, on which contributors rely to demonstrate the validity of their patches.
shujisado.org/2025/07/02/h...

Why Has DebConf Never Come to Japan? A Look at the Hurdles and Hopes DebConf is an annual developers’ conference organized by the Debian Project. Although it has a long history, it has never been held in Japan. Is that because Debian GNU/Linux is unpopular here? Not…

Last week, I discovered by chance that some Debian members are now aiming to bring DebConf to Japan. To raise public awareness, I published a brief article outlining the reasons previous bids were abandoned.
shujisado.org/2025/06/23/w...

Solving the Tree Swing Paradox: AI’s Final Boss Is the Human Client The term AGI is being trumpeted everywhere, but will every facet of human behavior really end up being replaced by AI? I remain skeptical.In software, for example, it already looks as though AI wil…

There is no doubt that AI will replace much of what humans do. But so long as we remain foolish, we will still insist on doing things ourselves.
shujisado.org/2025/06/20/s...

U.S. Copyright Office’s AI-Training Report amid Political Turbulence Last month, the U.S. Copyright Office released a study report on the fair-use implications of using copyrighted works without permission to train AI systems. As is widely known, the Register of Cop…

I drafted these comments only a few days after the report’s release but left them in my drafts folder when the unseemly political drama unfolded. I have now touched them up slightly and made them public as part of cleaning out that folder.
shujisado.org/2025/06/19/u...

When explaining the concept of open source, I often say, “We have to accept that Open Source software can also be used in warfare.” However, I never thought I would actually learn that it was used to discuss the operational plans of the world's most powerful military.

The Trump Administration Accidentally Texted Me Its War Plans U.S. national-security leaders included me in a group chat about upcoming military strikes in Yemen. I didn’t think it could be real. Then the bombs started falling.

It's really interesting.
I'm glad to know that high-ranking officials in the Trump administration were fans of Open Source apps.
www.theatlantic.com/politics/arc...

The Missing Piece in Japan’s Open Source Journey: Strong, Visionary Leadership Back in the 1990s, I was concerned that Japan was beginning to lag behind the United States in software technology. The primary reason for this delay was likely Japan’s heavy focus on hardware at the ...

In short, Japan needs strong leadership in the realm of open source. Of course, this isn’t limited to Open Source—it’s probably a fundamental issue for Japan as a whole.
www.linkedin.com/pulse/missin...

The Missing Piece in Japan’s Open Source Journey: Strong, Visionary Leadership Back in the 1990s, I was concerned that Japan was beginning to lag behind the United States in software technology. The primary reason for this delay was likely Japan’s heavy focus on hardware at the ...

Although more Japanese companies are gradually setting up OSPO, over the past few years I’ve felt that simply having an OSPO is not enough for JP companies to truly embrace Open Source. I intended to explain my thoughts here, but the explanation grew lengthy, so I wrote a blog post about it instead.

This article explores balancing our “freedom-first” philosophy with privacy/copyright law, Open Source community norms, and real AI/ML constraints. It also shows why some compromises on data requirements may be necessary right now.

Should ‘Open Source AI’ Mean Exposing All Training Data? DeepSeek has had a major global impact. This appears to stem not only from the emergence of a new force in China that threatens the dominance of major U.S. AI vendors, but also from the fact that t…

I joined the OSI’s early discussions on the Open Source AI Definition. From that experience, I wrote an article (personally) about requiring full training data in Open Source AI. If you’re curious about AI/ML openness, you might find it interesting.
shujisado.org/2025/02/18/s...

本稿は、自由の哲学を維持しながら、プライバシー法や著作権法、我々のコミュニティの慣習、あるいはAI/ML技術の現実との実用的なバランスを取る思考過程を解説するものである。言ってしまえば、まあデータ要件における妥協は現時点で何故必要であるかを説明するものとも言える。

基本的には完全なデータの公開を求める勢力と求めない勢力の対立であるわけだが、前者はDebian的な完全な自由を求める勢力が支持し、後者は巨大ベンダーやAIコミュニティの一部の支持が強い。現在のオープンソースAIの定義はその中庸的でありながら若干データ完全性を求めない所に線を引いている。

AIモデルがオープンソースであるために完全な学習データの公開は必要なのか? DeepSeekは世界に衝撃を与えているが、その要因としては、中国から米国の巨大AIベンダーを脅かす新たな勢力…

【完全保存版】DeepSeekのおかげでAIのオープンソース性について関心が高まっているが、AIがオープンソースであるための要件を考えると非常に面倒な問題もある。最たるものが学習データの完全な公開を必要とするかである。そこで、その点の世界で最も詳しい解説を書いた。
shujisado.com/2025/02/18/n...

“Torrenting from a corporate laptop doesn’t feel right”: Meta emails unsealed Meta’s alleged torrenting and seeding of pirated books complicates copyright case.

MetaがLlamaの学習へ海賊版データを利用したとされる件。倫理的にはアレとしてフェアユース抗弁が通るだろうなというのと、仮に違法としても既存モデルの契約としての有効性は残るのだろう。ただ、Llama契約の強力な伝播性から海賊版データが伝播という悪いイメージが付くわな
arstechnica.com/tech-policy/...

That may be one aspect, but underlying their actions is the influence of China’s domestic laws on generative AI. The Chinese government is simultaneously pursuing two seemingly contradictory objectives: powerful censorship and global collaboration.

It’s certainly true that using their generative AI service comes with risks, but at the same time, they distribute their model under the MIT license. Is this part of their business strategy?

DeepSeek in China: A Legal Overview of the 2023 Generative AI Regulations Since many people in the West have suddenly panicked about the possibility that the DeepSeek app is sending user information somewhere, I would like to offer a brief explanation. In 2023, China ena…

There seems to be a growing number of political messages claiming that “DeepSeek is dangerous,” stemming from the fact that DeepSeek Inc. sends user information to China through its generative AI service.
shujisado.org/2025/02/10/d...

Significant Risks in Using AI Models Governed by the Llama License Although it has already been explained that the Llama model and the Llama License (Llama Community License Agreement) do not, in any sense, qualify as Open Source, it bears noting that the Llama Li…

A detailed explanation of how the Llama License carries over to distilled models can be found in the article below:
shujisado.org/2025/01/27/s...

It remains unclear how Meta will respond. Engaging in litigation could entangle the issue with larger geopolitical concerns, while simply doing nothing would create a major loophole in the Llama ecosystem—posing a significant challenge for Meta.

In AI development, “distillation” constitutes an act of “using the outputs or results” under the Llama License. First, there is an obligation to prepend “Llama” to the name of model. Moreover, under contract-law principles, Llama License should be inherited by distilled model.

Among the DeepSeek R1 series, there is also a distilled model that uses Llama as the teacher model. On Hugging Face, that Llama-distilled model appears to be covered by the MIT License, but this is likely a contractual violation.

I expected abundant English material on Llama’s dangers but found surprisingly few references, so I translated my Japanese posts for a broader audience. Though written from a Japanese viewpoint, they should apply globally. Please share freely.

One article explains why Llama isn’t truly Open Source. The other highlights serious license risks, including how the Acceptable Use Policy can impose broader constraints than conventional copyleft. Originally aimed at Japanese businesses, both were well received.

The Hidden Traps in Meta’s Llama License — An Explanation of Llama’s Supposed “Open Source” Status and the Serious Risks of Using Models under the Llama License — It is widely recognized—despite Meta’s CEO persistently promoting the…

I’ve long said “Llama isn’t Open Source and is actually dangerous.” Because I'm seen as an OS advocate, many thought I merely disliked Llama’s non-OS status. Worried the real dangers were overlooked, I published two articles addressing those risks.
shujisado.org/2025/01/27/t...

特に直接AIモデルを使用しないサービスのエンドユーザーまでMeta社の利用規約を遵守義務があり、そこへ規約変更の権利が組み合わさると、最近のVISA決済停止問題のような日米の倫理観等の違いからくる問題が発生する要因になるのではないか。これを本邦の派生モデル企業が注意しているかは分からない。

7億MAU制限は単に範囲が広いと思われるだけだろうが、Metaに随時の変更権限を持たせる組込型利用規約の存在、その利用規約の仕組みを使っての敵対勢力の排除の可能性、コピーレフトを超える強い契約の伝播性といった性質は従来のソフトウェアの使用許諾を大幅に超えてMeta社の権限を拡張するものと思う

Llamaライセンス契約を適用するAIモデルを使用する際の多大なリスク Meta Platforms社のLlamaモデルならびにLlamaライセンス契約(Llama Communit…

Llama第二弾というより本論公開。AIモデルとしてもシステムとしてもオープンソースではないと前回書いたが、今回は派生を含めてLlamaをAIモデルとして使用することのリスクを列挙している。本邦の派生モデル開発企業だけでなくユーザー企業へも罠が幾つもあると考えている。
shujisado.com/2025/01/20/l...

なお、単なる使用許諾契約としてのLlamaの危険性はもっと分厚く解説を書いている。特に本邦企業にとっては危険だろうと思うので、オープンソース性解説より丁寧に書く。(なのでもうちょいかかる)

オープンソース性について詐欺師界隈に指摘しても「厳密にはオープンソースではないが」という枕詞をつけてお茶を濁す者がいるが、この記事では厳密どころか最初から最後まで全く適合性がないことを書いた。通常、OSI承認プロセスでもここまで説明しないが、書くべき状況であると判断した。