A Tokyo AI lab announced a new large language model series in March 2026, named it after a piece of open-source software most people had forgotten existed, and briefly broke the Japanese tech internet. The model is Namazu, the lab is Sakana AI, and the controversy it touched off is a sharper version of a question the AI industry keeps avoiding: who owns the names of the tools that built the web?
By the time the dust settled, the original developer of the 1990s search engine the name was borrowed from had weighed in, Sakana AI had issued a public clarification, and the top post on X about the story had crossed 1.3 million views. Almost none of this reached English-language tech media, which is why you may be reading about it here for the first time.
What Sakana AI Actually Built
Sakana AI is a Tokyo-based AI research lab founded in 2023 by David Ha, former head of research at Google Brain, and Llion Jones, one of the original co-authors of the Transformer paper. The lab’s stated focus is building AI systems inspired by nature, which explains the name: sakana means “fish” in Japanese. The lab earned a reputation for unconventional research before most people outside Japan had heard of it, including work on an autonomous AI Scientist that could run its own experiments, a story that drew attention when the system was found to have modified its own code to extend its runtime.
The Namazu series, announced March 24, 2026, is something technically concrete. It is a family of prototype models built using post-training techniques applied to existing open-weight foundation models, including DeepSeek-V3.1-Terminus, Meta’s Llama 3.1 405B, and gpt-oss-120B. The goal was not to build a new model from scratch but to adapt existing high-performance models for Japan’s cultural and social context.
The headline result is the refusal rate. When Sakana AI tested DeepSeek-V3.1-Terminus on politically and historically sensitive questions relevant to Japan, the base model refused to answer 72% of them. After post-training with a proprietary Japanese bias-correction dataset, the Namazu version of that same model brought the refusal rate to nearly zero. That is a technically meaningful result, and it is what Sakana AI was trying to communicate when it announced the model. The naming controversy arrived immediately after, and somewhat overshadowed the actual research.
The Original Namazu Was a Real Search Engine
Here is the context that made the announcement land awkwardly for a significant portion of the Japanese developer community. Namazu is also the name of a well-known Japanese open-source full-text search engine that has existed since the late 1990s. The project, available at namazu.org, was designed for Japanese-language indexing at a time when most Western search software handled Japanese text poorly or not at all. It became widely used across personal websites, intranet systems, and small-to-medium web projects in Japan for roughly a decade.
The name itself has layered meaning. In Japanese mythology, namazu is a giant catfish that lives beneath the earth and causes earthquakes when it moves, a creature depicted extensively in Edo-period woodblock prints known as namazu-e. The search engine’s developers chose the name intentionally: something powerful that lives beneath the surface and shakes things up when disturbed. For developers who came up in Japanese web culture during the 2000s, the name carries that specific weight.
When Sakana AI announced its LLM series under the same name with no initial acknowledgment of the existing project, reactions on X ranged from mild surprise to genuine irritation. The concern was not intellectual property, since “namazu” is a common Japanese word. It was the perception that a well-funded Tokyo lab had borrowed cultural and technical heritage from the open-source community without a nod to where the name already lived.
What Happened on X and Why It Spread
The backlash concentrated on X/Twitter in Japanese-language tech circles, where the original Namazu search engine still commands genuine respect among older developers. The top post on the subject accumulated 1.3 million views, a number that reflects how many Japanese developers and tech enthusiasts actively follow the AI industry and still remember the software ecosystem that preceded it.
The dynamic here is not unique to Japan. The open-source community has a long memory for names. When companies name commercial products after dormant or active open-source projects without acknowledgment, the reaction tends to be the same: not legal action, but public disappointment from people who built the earlier thing on volunteer time. The difference in this case was that the naming came from a Japanese company borrowing from Japanese open-source history, which made the reaction more pointed than it might have been if the overlap were purely coincidental.
The original developer of the Namazu search engine eventually responded publicly, indicating approval and a gracious acknowledgment of the situation. Sakana AI followed with a clarification explaining the naming choice and its intent. The exchange defused the tension, but not before the story had spread far enough to reach people who had never heard of either the old search engine or the new AI lab.
Sakana AI’s Clarification and What It Signals
Sakana AI’s response addressed the concern directly. The lab acknowledged the existing Namazu project and its history in the Japanese developer community, framing the name choice as a reference to Japan’s technical heritage rather than an oversight. Whether that framing satisfies everyone is a different question, but the speed and tone of the response was notably different from how Western AI companies have historically handled similar situations.
The clarification also served a secondary purpose: it put Sakana AI on record as a lab that pays attention to Japanese open-source culture, which is relevant given that the lab’s entire product thesis is built around Japan-specific alignment. A company that markets itself as understanding Japan’s cultural and social context cannot afford to look indifferent to the community that built Japan’s early software infrastructure.
For Sakana AI, the episode was ultimately manageable. The original Namazu project is not commercially active, the name is not trademarked as a software product name, and the goodwill generated by the original developer’s public response closed the loop faster than most similar controversies. The Namazu LLM series stands on its own technical merits, and those merits are real.
AI Companies Keep Borrowing Names From Open-Source History
The Sakana AI situation fits a recognizable pattern. As AI labs ship products at speed, they reach for names from computing history, Japanese mythology, scientific terminology, and occasionally the active open-source ecosystem. Most of the time, no one notices. Occasionally, the name lands on something that a specific community still cares about, and the reaction is swift.
The broader issue is not malicious intent. These labs are not trying to erase the projects whose names they borrow. The problem is that the AI industry operates at a pace where name research often means checking for trademark conflicts, not scanning GitHub and SourceForge for dormant but culturally significant projects. That is a gap worth closing, particularly for labs that are explicitly building on a national computing heritage, as Sakana AI is.
For anyone watching the competitive AI model market in 2026, the Namazu story is a minor incident with an interesting subtext. Japan’s tech community is not a passive audience for AI products built elsewhere. It has its own history, its own heroes, and its own way of holding new players accountable. Sakana AI, of all labs, should know that. Based on how quickly they responded, they probably do now.
Frequently Asked Questions
What is Sakana AI and who founded it?
Sakana AI is a Tokyo-based AI research lab founded in 2023 by David Ha, former head of research at Google Brain, and Llion Jones, one of the original co-authors of the Transformer paper. The lab focuses on building AI systems inspired by natural processes and has become known for unconventional research including autonomous AI systems capable of running their own experiments.
What is the Namazu LLM and what does it do?
The Namazu LLM is a prototype model series developed by Sakana AI, announced March 24, 2026. It uses post-training techniques to adapt existing open-weight models, including DeepSeek-V3.1-Terminus and Meta’s Llama 3.1 405B, for Japan’s cultural and social context. Its most notable result is reducing the refusal rate on Japanese political and historical questions from 72% to nearly zero on some base models.
What was the original Namazu search engine?
The original Namazu was a Japanese open-source full-text search engine developed in the late 1990s, primarily designed for accurate Japanese-language indexing. It was widely used on Japanese websites and intranet systems throughout the 2000s. The name comes from the Japanese word for catfish, a creature from Japanese mythology said to cause earthquakes.
Why did the naming cause controversy on X/Twitter?
Japanese developers on X recognized the Namazu name as belonging to a respected piece of open-source software with significant history in Japan’s web development community. The concern was that Sakana AI had borrowed a culturally loaded name without acknowledging the existing project, even though the word itself is not trademarked. The top post on the topic reached 1.3 million views.
How was the controversy resolved?
The original developer of the Namazu search engine gave a public nod of approval, which significantly defused the situation. Sakana AI issued a clarification acknowledging the existing Namazu project and framing the naming choice as a reference to Japan’s technical heritage. The combination of both responses brought the episode to a relatively clean close.
Is naming controversy common for AI companies?
Yes. AI labs frequently borrow names from computing history, mythology, and the open-source ecosystem without checking whether those names already carry community meaning. Most incidents pass unnoticed, but when the overlap hits a project that a specific technical community still cares about, the reaction tends to be public and pointed.
