AI Daily News — 2026-10-07 (Wed) Evening News

[Overview] The biggest topic is that OpenAI has published 377 new mathematical results on GitHub, and that AI's mathematical ability has entered a new level. At the same time, AI-induced restructuring has spread to banking and finance, such as HSBC's 70% cut in advisors in its UK wealth division and 15% cut in FICO. xAI is releasing Grok 4.5 as SpaceXAI, and SpaceX is seeking $40 billion to buy Nvidia chips.

[Anthropic / Claude trends] The Department of Defense (DOD) has suspended its use of Anthropic AI after a court upheld its supply chain risk designation. Anthropic's position in government procurement has suffered a major setback. It was also reported that Microsoft and Meta are reducing their internal use of Claude in favor of in-house tools. Even large private customers are under increasing pressure to produce their products in-house. ClawSecure's testing reportedly contradicted Anthropic's safety claims, a blowback for brands that tout safety. CEO Dario Amodei's 2025 compensation was $18 million, which was in the mid-range for tech CEOs. We also observed the theft of credentials and MFA codes through fake ChatGPT, Gemini, and Claude advertising portals, continuing to pose risks to users.

[Trends in OpenAI] OpenAI has published 377 mathematical results on GitHub and has also published an explanatory article. The Verge and WSJ report that ``AI that made a fuss by solving one problem went on to solve hundreds more.'' However, Gizmodo is sarcastic about the way it was published and the tone of the article, and the verification method and evaluation of reproducibility are future issues. For developers, the company announced the Decisions API and focused on agent competition at DevDay 2026. It received criticism from Wikimedia over the cost of data scraping. The apology visit to Australia has been criticized for leaving key questions unanswered. In the medical field, ChatGPT's efforts to improve the quality of health consultations were also introduced.

[Google AI trends] Flash and Pro will no longer be available in the free version of Gemini on October 9th. It appears that the aim is to change the experience of free users and encourage them to switch to paying. Sundar Pichai announced EmbeddingGemma 2, the first open and natively multimodal embedding model. Meanwhile, GOOGL stock fell on concerns about an outflow of AI talent, but Pichai and analysts downplayed the concerns, and the stock rebounded in the pre-market. Adoption in peripheral services is also expanding, such as Ant International's Antom compatible with Gemini and expanding into Asia.

[Trends of other AI companies] xAI has been renamed SpaceXAI and released Grok 4.5, and there is also a proposal to rename it to "SpaceXSI". Grok was also reported to have problems. Meta is promoting in-house production by introducing its own AI "Muse", and Citi expects the scale to be $27 billion in 2030. Sierra, Meta, Genesys, and NiCE have joined the open standard for personal AI agents. Mistral has announced the unreleased model "Le Chonk", and its revival is attracting attention. DeepSeek reportedly plans to double its latest funding to $15 billion. On the other hand, an article stating that the company and Moonshot AI are planning an IPO states that it will take place in 2025, and the consistency of the timing has not been confirmed. DeepSeek V4.1 Flash's Q4 quantization requires 294GB of RAM, which is more difficult to run locally compared to Q2's 152GB. Ivo has released an open source DeepSeek model for contracts.

[AI celebrities' statements and trends] - Sam Altman (OpenAI CEO): "The world should accept 'some bad things' because the benefits of AI outweigh the harms," he said, drawing a line with Anthropic over AI regulation. This has sparked a strong backlash, calling it ``insane.'' - Elon Musk (xAI founder/Tesla CEO): TSMC has said it will not operate the Terafab AI chip complex. SpaceX is strengthening its relationship with Nvidia, seeking $40 billion to buy AI chips. He also expressed support for Trump's name change proposal, saying, "There is no more AI." Citing Jensen Huang's deepfake video, he poked fun at RAM prices. - Jensen Huang (NVIDIA CEO): Nvidia's market capitalization reached a record high of $5.65 trillion. A deepfake video of Huang went viral, but it's not his own words. - Dario Amodei (Anthropic CEO): Compensation in 2025 was $18 million. At the same time, there are reports that company officials have acknowledged that the company is inadequately prepared for the catastrophic risks of AI, but the sources are secondary and the details have not been confirmed. - Sundar Pichai (Google/Alphabet CEO): Announced EmbeddingGemma 2. The company downplays concerns about a loss of human resources. - Mark Zuckerberg (Meta CEO): It was reported that the company will launch Muse and move towards in-house production. In the debate over AI regulation, it has been pointed out that there is a risk that the US will cede control to China and the EU. - Geoffrey Hinton (AI Godfather / Nobel Laureate): He expressed his regrets about inventing AI and called for an FDA-type approval system to be established before its release. - Yann LeCun (Meta Chief AI Scientist): This article has little relevance, and I cannot confirm his new statements. - Demis Hassabis, Ilya Sutskever, Daniela Amodei, and others: New statements by the individuals could not be confirmed. Articles related to the movie "Artificial" focus on the depiction of characters, including Sutskever.

[Breakthroughs and innovations in AI technology] The biggest topic is OpenAI's 377 mathematical results. Peer review of the correctness and novelty of each proof has not yet been completed, and the evaluation has not been finalized. Valeo demonstrated the reach of aluminum cooling technology and durability at OCP Global Summit 2026 to meet AI demand in data centers. Cooling and power are increasingly important physical constraints for AI calculations.

[AI Regulation/Policy/Social Impact] Altman's ``accept bad'' statement and Hinton's proposed approval system are polar opposites of regulatory theory. The hidden costs of federal preemption of state laws were also discussed. Although the UK claims to be a leader in safety, the starting point has not yet been determined. New Mexico's attorney general and state legislators have come up with regulations for runaway AI models. The court determined that attorney-client privilege does not extend to a client's interactions with an AI. In M&A, it has been pointed out that there is a risk of inheriting the acquired company's AI responsibilities due to insurance holes. The head of the IMF has warned that while AI will boost growth, it will also fuel inflation. On the employment front, Amazon and other companies are increasing their ``boomerang hiring'' efforts to rehire retired employees as part of a shake-up of AI restructuring.

[Integrated analysis: What is happening now] In terms of power, there is a noticeable trend among platform companies to reduce their dependence on external models. Microsoft and Meta are reducing Claude internally in favor of their own tools. Anthropic is also facing some government headwinds from the DOD designation. Meanwhile, OpenAI maintains a technological presence with its mathematical results and API (Decisions API), and faces criticism. xAI has teamed up with SpaceX and is looking to secure Nvidia chips for a huge amount of money. Nvidia's market capitalization of $5.65 trillion indicates that demand for computing resources remains strong. On the technical side, progress is being made in parallel in the mathematical demonstration of inference capabilities, the development of agent standards, open embedding models, and the amount of memory required for local execution. On the investment side, reports of an expansion in the scale of DeepSeek's procurement and developments in semiconductor manufacturing (TSMC and Terafab) are notable. In geopolitics, there are concerns that the United States may cede regulatory leadership to China and the EU. On the social front, while personnel reductions due to the introduction of AI have spread to the financial sector, there has also been a reversal due to rehiring. Please note that some articles, such as the timing of the DeepSeek IPO, are inconsistent in date and content, and should be treated with caution.

[Future points of interest] 1. Will OpenAI's 377 mathematical results be evaluated as appropriate after peer review by external mathematicians? 2. To what extent will the DOD designation affect Anthropic's business for governments and large corporations? 3. Will SpaceX's $40 billion chip raise and DeepSeek's funding scale materialize?