🗾ITmedia AI+ (日本)•較早收集於 34m
軟銀子公司AI企業與NICT共同研究實現「安全AI」
💡SoftBank-NICT safe AI research may shape industry safety protocols
⚡ 30-Second TL;DR
有什麼變化
軟銀AI研發子公司SB Intuitions與NICT展開共同研究
為什麼重要
此合作可能加速日本安全AI標準，影響企業全球AI倫理與部署實務。
下一步行動
Follow NICT's research portal for emerging safe AI frameworks and publications.
誰應關注：Researchers & Academics
關鍵要點
- •軟銀AI研發子公司SB Intuitions與NICT展開共同研究
- •主要目標為開發安全AI解決方案
- •公告強調AI安全進展的重點
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 9 個來源。
🔑 增強重點摘要
- •SB Intuitions develops the 'Sarashina' series of Japanese-specialized large language models, including Sarashina2-70B and the multimodal Sarashina 2-Vision, leveraging SoftBank's AI computing infrastructure[2][3][5].
- •NICT established the CREATE lab in 2024-2025, specializing in AI security research including model interpretability, cybersecurity automation, and US-Japan collaborations with partners like NIST and MITRE[1].
- •SB Intuitions has pursued AI safety through its Sarashina Lab, prioritizing safety, reliability, and ease of use in advanced AI systems[8].
🔮 前景展望AI analysis grounded in cited sources
This partnership will advance Japanese-specific AI safety benchmarks by 2027
Combining SB Intuitions' Sarashina models with NICT's CREATE focus on interpretability and risk assessment leverages Japan's domestic data security strengths for specialized safe AI development.
Enhanced US-Japan AI security ties via NICT will integrate SoftBank models into global standards
NICT's CREATE expansions with US partners like NIST provide a framework for testing and sharing SoftBank's safe AI technologies internationally.
⏳ 時間線
2023-03
SB Intuitions established as SoftBank subsidiary for Japanese LLM R&D
2023-08
SoftBank launches SB Intuitions to develop homegrown LLMs
2024-06
SB Intuitions releases initial Sarashina models (7B, 13B, 65B parameters)
2024-11
SB Intuitions releases 400B parameter Japanese LLM
2025-01
SB Intuitions partners with Chugai and SoftBank for generative AI in clinical development
2025-06
SB Intuitions unveils Sarashina logo and advances multimodal models
📎 來源 (9)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- spfusa.org — AI Assurance Roundtable Event Summary
- sbintuitions.co.jp — Company
- dentsu.co.jp — 0925 010948
- sbintuitions.co.jp — En
- sbintuitions.co.jp — News
- softbank.jp — Future Tech
- csri.nict.go.jp — 260108 01
- sbintuitions.co.jp — Sarashinalab
- internationalaisafetyreport.org — International AI Safety Report 2026
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： ITmedia AI+ (日本) ↗
每週 AI 簡報
每週一封，可隨時退訂。