Skip to main content

Rakuten GroupRakuten marketed 'homegrown' AI 3.0 model that was a rebranded DeepSeek-V3, after deleting the open-source license file

In March 2026, Rakuten released 'Rakuten AI 3.0,' promoted as Japan's most powerful domestically developed AI model. Independent technical analysis (published on GitHub and picked up by multiple Asia-tech outlets) showed the model's architecture and weights were derived from Chinese lab DeepSeek's open-source DeepSeek-V3, and that Rakuten had removed DeepSeek's MIT license attribution file from its public repository, obscuring the model's true origin. The disclosure prompted backlash in Japan, particularly given Rakuten's positioning of the model as a domestic achievement while it may have been eligible for government AI development subsidies. After the removal was discovered and criticized publicly, Rakuten restored a NOTICE file containing DeepSeek-V3's original license statement, bringing the release back into technical compliance with the MIT license, but did not fully address criticism about the initial lack of transparency regarding the model's foundation.

Scoring Impact

TopicDirectionRelevanceContribution
Corporate Transparency-againstsecondary-0.50
Open Source Licensing Integrity-againstprimary-1.00
Overall incident score =-0.409

Score = avg(topic contributions) × significance (medium ×1) × confidence (0.55)

Evidence (1 signal)

Confirms product_decision Mar 18, 2026 reported

Technical analysis shows Rakuten AI 3.0 is a rebrand of DeepSeek-V3 with deleted license attribution

A GitHub technical investigation (hama-jp/RakutenAI_Report) and multiple Asia-tech outlets reported that Rakuten AI 3.0's architecture matched DeepSeek-V3 and that Rakuten had removed the required MIT license file crediting DeepSeek, restoring it only after the discovery went public.

Related: Same Topics