Daily AI Search Memo (2025/12/2 Issue)
Update Date: 2025/12/2
Executive Summary
In terms of policy, Hungary, Australia, India, and the United States are each advancing AI safety institutions and domestic legislation, further strengthening the trend of AI Safety Institute networks + EU AI Act implementation.
In Japan, The Mainichi Newspapers and Sankei Shimbun protested against the generative AI search engine Perplexity for copyright infringement, bringing the conflict between media and generative AI companies to the surface.
Economically, while large-scale investments in frontier AI companies and foundation models continue, such as the HSBC x Mistral AI partnership and Black Forest Labs' $300 million funding round, Japanese companies are also beginning to implement AI for expense settlement and regional information services.
In the smartphone and semiconductor sectors, ByteDance's 'Doubao' and the TPU collaboration between MediaTek and Google are accelerating on-device inference and AI chip integration.
Socially, in addition to the criticism of Perplexity by Japanese media, new cybersecurity risks and the overheating of the AI talent market have been reported, highlighting the impact of AI adoption on labor, safety, and rights protection.
Technologically, multi-agent collaboration (Fujitsu), agent-based call centers (Amazon Connect), the math-specialized model DeepSeek-Math V2, and on-device multimodal assistants (Doubao) show that 'agentification,' 'efficiency,' and 'specialization' are common keywords.
Politics Analysis
1. Hungary implements domestic law for the EU AI Act
Source: https://datalawgy.substack.com/p/what-is-new-in-data-technology-and-966 (DataLawgy / 2025-12-01)
Key Point: The Hungarian Parliament adopted a law to apply the EU AI Act domestically, which came into effect on December 1st. Compliance checklists based on the definition of generative AI and risk classification have been established.
Impact: As domestic implementation of the EU AI Act progresses, a common compliance foundation is expanding for AI providers and users within the region. This indicates that an operational phase at the practical level has arrived for Eastern European countries as well.
2. Australian government officially launches AI Safety Institute (AISI)
Source: https://tickernews.co/cybersecurity-concerns-rise-amidst-ai-industrial-adoption/ (Ticker News / 2025-12-01)
Key Point: The Australian government has launched a new AI Safety Institute (AISI) aimed at managing risks associated with AI adoption in the industry. It will conduct risk assessments and develop safety guidelines at the national level.
Impact: Public oversight and verification frameworks for AI adoption in infrastructure and critical industries will be strengthened. Coordination with cybersecurity and critical infrastructure guidelines will be a particular focus.
3. Indian government concretizes IndiaAI framework including AI Safety Institute
Source: https://www.thehindubusinessline.com/catalyst/daggers-drawn-between-ai-evangelists-and-doomsayers/article70339626.ece (The Hindu Business Line / 2025-11-30)
Key Point: The Indian Ministry of Electronics and Information Technology is considering a concrete framework for the IndiaAI mission, including the establishment of an "AI Safety Institute" to lead the safe implementation and regulation of AI.
Impact: A move to simultaneously develop sovereign AI infrastructure and safety regulations in India, a country with a large population and multiple languages. Connection with international AI safety networks will be a future focus.
4. The Mainichi Newspapers protests against generative AI search engine Perplexity for copyright infringement
Source: https://www.mainichi.co.jp/info/20251201.html (The Mainichi Newspapers / 2025-12-01)
Key Point: The Mainichi Newspapers sent a letter of protest to US-based Perplexity, alleging that the company is collecting and using its articles without permission. It pointed out the possibility of a violation of the Unfair Competition Prevention Act and demanded the cessation of unauthorized use and the commencement of discussions.
Impact: This is a case where a major Japanese news organization has formally raised objections to a generative AI search service from the perspective of copyright and unfair competition, which could influence the interpretation of "fair use" in generative AI training and RAG.
Economics Analysis
1. HSBC enters into a multi-year partnership with Mistral AI to implement generative AI across the company
Source: https://www.hsbc.com/news-and-views/news/media-releases/2025/hsbc-and-mistral-ai-join-forces-to-accelerate-ai-adoption-across-global-bank (HSBC Holdings / 2025-12-01)
Key Point: Major British financial institution HSBC has signed a multi-year contract with French company Mistral AI to host Mistral's commercial models on its own infrastructure, deploying generative AI across the bank's operations, including financial analysis and risk assessment.
Impact: This is a case where a regulated industry, banking, is moving forward with the adoption of self-hosted foundation models rather than relying on cloud APIs, symbolizing large-scale deployment that prioritizes data sovereignty and compliance.
2. ByteDance launches "Doubao" in the Chinese smartphone market
Source: https://www.reuters.com/world/china/bytedance-rolls-out-ai-voice-assistant-chinese-smartphones-2025-12-01/ (Reuters / 2025-12-01)
Key Point: ByteDance has released an AI voice assistant powered by its own foundation model, "Doubao," and integrated it at the system level into smartphones such as those from ZTE. It allows for voice-based app operation and payment processing.
Impact: The construction of an on-device AI ecosystem within China is progressing, and competition is intensifying not only at the app platform level but also at the OS layer. There are concerns that this will strengthen platform dominance by involving payments and commerce.
3. Black Forest Labs raises $300 million for image generation model FLUX
Source: https://www.globenewswire.com/news-release/2025/12/01/3196629/0/en/Black-Forest-Labs-Announces-Series-B-Investment-to-Accelerate-Frontier-Visual-Intelligence.html (GlobeNewswire / 2025-12-01)
Key Point: Black Forest Labs, the developer of the image generation model "FLUX," has raised $300 million in a Series B round led by AMP and Salesforce Ventures, reaching a valuation of $3.25 billion.
Impact: This is a large capital injection into a European-based generative image startup, indicating continued expectations for players with strengths in text fidelity and high-quality rendering. It increases the investment capacity for video generation and open-source development.
4. RAKUS launches beta version of "RAKURAKU AI Agent for RAKURAKU Seisan"
Source: https://www.rakus.co.jp/news/2025/1201.html (RAKUS Co., Ltd. / 2025-12-01)
Key Point: IT service company RAKUS has added an AI feature, "RAKURAKU AI Agent," to its expense reimbursement cloud service "RAKURAKU Seisan" as a beta version. The AI analyzes receipt data and application history to automatically generate expense reports.
Impact: The automation of voucher entry and verification using generative AI is becoming full-scale even in the back-office operations of Japanese companies. This is a functional enhancement that directly addresses the needs of small and medium-sized enterprises to reduce indirect operational costs.
5. Shimotsuke Shimbun begins offering "Shimotsuke Shimbun Generative AI" for corporate use
Source: https://prtimes.jp/main/html/rd/p/000000033.000077069.html (Shimotsuke Shimbun Co., Ltd. / 2025-12-01)
Key Point: Tochigi Prefecture's regional newspaper, Shimotsuke Shimbun, launched a corporate generative AI service on December 1st that utilizes 15 years of its own article data. It is available from 30,000 yen per month, and users can choose from the latest models such as GPT and Gemini.
Impact: This is a move by a regional newspaper to commercialize its archives as a knowledge-based generative AI for local businesses. It serves as a demonstration case for a new revenue stream through the combination of local data and general-purpose models.
Social Analysis
1. Sankei Shimbun files formal protest against unauthorized use of articles by Perplexity
Source: https://www.sankei.jp/wp-content/uploads/2025/12/2025.12.01-Protest-Regarding-Unauthorized-Use-of-Our-Articles.pdf (Sankei Shimbun Co., Ltd. / 2025-12-01)
Key Point: Sankei Shimbun has sent a protest letter to the US generative AI company Perplexity, demanding the cessation of use and deletion of its articles based on copyright infringement. It severely criticized the company for "free misappropriation" of Sankei News in its answer generation.
Impact: The conflict between news organizations and generative AI services regarding the legitimacy of scraping and RAG usage is becoming apparent. Although Japan was considered tolerant of AI training, a trend toward strengthening rights protection, centered on news media, is gaining momentum.
2. Director James Cameron expresses concern, calling generative AI "horrifying"
Source: https://techcrunch.com/2025/11/30/avatar-director-james-cameron-says-generative-ai-is-horrifying/ (TechCrunch / 2025-11-30 (Reported 12/1 JST))
Key Point: Director James Cameron, known for "Avatar" and other films, stated in an interview that the risk of creators being replaced by generative AI is "horrifying," emphasizing the importance of human creativity and the protection of rights.
Impact: As a concern expressed by a prominent creator representing the film industry, this may accelerate discussions on AI usage rules and compensation distribution in scriptwriting and video production settings.
3. Rising cybersecurity concerns accompanying the acceleration of AI adoption in industry
Source: https://tickernews.co/cybersecurity-concerns-rise-amidst-ai-industrial-adoption/amp/ (Ticker News / 2025-12-01)
Key Point: As the introduction of autonomous AI agents and industrial AI systems progresses in the industrial sector, cybersecurity experts are sounding the alarm regarding the expansion of new attack surfaces.
Impact: The risk of AI embedded in control systems and supply chains becoming targets of attack is increasing, making 'secure-by-design'—incorporating security and safety from the design phase of industrial AI—increasingly essential.
Technology Analysis
1. Amazon Connect announces agent AI features for call centers
Source: https://www.aboutamazon.com/news/aws/aws-re-invent-2025-ai-news-updates (Amazon News / 2025-11-30 (Dec 1 in JST))
Key Point: At the AWS 're:Invent 2025' event, it was announced that agent-based AI features will be integrated into the call center service Amazon Connect. Using the 'Nova Sonic' AI model, it understands and reasons through customer conversations to automate voice and message responses.
Impact: A division of labor where AI handles primary call center responses while human operators focus on complex cases is becoming a realistic option. This is an important case study for the integration of speech recognition, natural language understanding, and dialogue control.
2. Fujitsu develops multi-AI agent collaboration technology for cross-company use
Source: https://global.fujitsu/ja-jp/pr/news/2025/12/01-02 (Fujitsu / 2025-12-01)
Key Point: Fujitsu has developed 'multi-AI agent collaboration technology' that allows AI agents belonging to different companies to work together to optimize the entire supply chain. Demonstration experiments will begin in January 2026 with Rohto Pharmaceutical and Tokyo University of Science.
Impact: Even when information shared between companies is limited, an architecture where each company's AI agent issues instructions and negotiates to achieve overall optimization is considered effective for supply chain operations under incomplete information, such as during disasters.
3. DeepSeek releases self-verifying mathematical reasoning model 'DeepSeek-Math V2'
Source: https://gigazine.net/gsc_news/en/20251128-deepseek-math-v2/ (Gigazine / 2025-11-30 commentary)
Key Point: China's DeepSeek has announced 'DeepSeek-Math V2,' a model specialized in mathematics. By scaling verification compute and automatically labeling difficult proofs, it introduces a self-verification mechanism that allows the model to improve its own reasoning capabilities.
Impact: It is said to have achieved gold-medal-level performance at the International Mathematical Olympiad (IMO), indicating a direction for 'self-improving AI' in fields where rigorous verification is possible, such as mathematics. This has significant implications for solving high-difficulty problems in research and education.
4. ByteDance's 'Doubao' achieves on-device multimodal operation
Source: https://www.moomoo.com/news/post/62231396/bytedance-stirs-the-on-device-ai-trend-doubao-releases-a (Moomoo News / 2025-11-30)
Key Point: It has been reported that ByteDance's AI assistant 'Doubao' has achieved on-device multimodal capabilities, allowing it to understand information on the screen and perform tasks such as photo editing or cross-app order processing using only voice commands.
Impact: 'Agent-like' operations on smartphones (screen recognition + cross-app interaction) are nearing practical application, serving as a prime example of on-device AI that improves user experience while reducing reliance on the cloud.
5. AWS releases Agentic AI workflows for the media industry
Source: https://aws.amazon.com/blogs/media/agentic-ai-headlines-aws-reinvent-2025-across-media-and-entertainment/ (AWS Media Blog / 2025-11-30 (Dec 1 in JST))
Key Point: In conjunction with re:Invent, AWS announced Agentic AI workflows for the advertising and media industry. An architecture was presented that autonomously handles tasks such as automatic generation of video highlights and complex metadata management.
Impact: The design assumes that agents will continuously execute parts of the creative production and distribution flow, suggesting that the standard workflow in the media industry may shift to an 'agent-first' model.
Comprehensive Analysis
Regarding policy and regulation, the AISI and the domestic implementation of the EU AI Act indicate that safety research institutions and rule-making are progressing in various countries, with international cooperation and competition surrounding AI safety occurring simultaneously.
Economically, major financial institutions, startups, and local companies are implementing generative AI at their respective scales, with diverse implementation patterns emerging, ranging from cloud API dependency to self-hosted and on-device models.
Socially, copyright disputes between media outlets and generative AI companies as well as impacts on cybersecurity and the labor market are becoming apparent, making the design of a balance between 'AI benefits' and 'rights and safety' a focal point.
Technically, the shift toward agents (in call centers, supply chains, media, and smartphones) and specialization in expert domains like mathematics is advancing, with the structure of 'general-purpose model + specialized workflow' emerging as a common denominator.
Points to Watch in the Future
To what extent will AI safety institutes and consortia in various countries release concrete evaluation criteria and benchmarks?
How will the developments surrounding Japanese media companies and Perplexity ripple out into lawsuits, licensing agreements, and industry-wide rule-making?
How will cases like HSBC and Black Forest Labs influence generative AI investment and adoption decisions in other industries and regions?
To what extent will agentic AI, such as Amazon Connect, Fujitsu's multi-agents, and Doubao's on-device features, become embedded in actual business and daily life?
