🔍 Advertising Regulation in Japan (Promotional Material Review) and Artificial Intelligence JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes

AI Daily News Monthly Report — 2026-07

Monthly AggregatePast 30 days meta-analysis + cross-period review
🗓️ Monthly
Source: Google News RSS / official blogs  ·  2026-07  ·  1811
AI 統合分析 / AI INTEGRATED ANALYSIS2026-07 () — 🗓️ Monthly ·
I understand.

I understand. Create a meta-analysis report based on all AI news for 30 days.

AI industry meta-analysis report (July 2026 period)

[Summary] Most important trends 3-5 points

First, the incident in which Google restricted the provision of the Gemini API to Meta due to a lack of computing resources made visible the structural problem of supply constraints in the AI industry. It was also revealed that Meta relied on Google AI for customer service, advertising tools, and content moderation, exposing the risks of over-reliance on its own model. Second, the export control issue surrounding Anthropic's Claude Mythos 5/Fable 5 took an unusual turn, with the U.S. government temporarily banning and then lifting the ban on the model, indicating that the Frontier model had begun to be treated as a de facto strategic material. Third, xAI's Grok 4.5 (1.5 trillion parameters V9) has entered private testing within Tesla and SpaceX, and Musk has declared monthly model releases, further accelerating the race for development speed. Fourth, DeepSeek's inference acceleration framework DSpark (up to 85% faster) showed its superiority over American competitors in the practical performance of a Chinese open source model. Fifth, Anthropic's introduction of Claude Science and the expansion of its partnership with the California state government indicate that Frontier Lab's business area is expanding into the research and administrative fields.

Anthropic / Claude trends

It announced a workbench for scientific research called Claude Science, and adopted a strategy of differentiating itself by providing workflows rather than introducing new models. At the same time, the company has started its own drug discovery program, marking a full-scale entry into the biopharmaceutical field. Under Governor Newsom, we have partnered with the California state government to provide AI tools, and have entered into a contract to provide half-price AI tools to state employees. Meanwhile, the U.S. administration placed temporary release restrictions on Claude Mythos' cyber capabilities, which were later lifted for limited organizations. Fable 5 is also back after a two-week ban. In terms of human resources, it was reported that Google's Gemini team was poached by Anthropic, and in its relationship with Amazon, rising costs have led Amazon to consider options for the OpenAI/Nova model. Performance evaluation of Opus 4.8 also became a hot topic.

OpenAI trends

A preview of GPT-5.6 (codename Sol) was announced, but the official release was postponed due to a US government review request. Frontier strategic partnership with HP to advance AI deployment across the enterprise, and announced a national security partnership with Booz Allen. In terms of cost, it has been reported that they have discovered a new method that reduces inference costs by half. Meanwhile, the company faces new lawsuits over the FSU shooting, alleging that the chatbot was brought to market too quickly. Analysts have pointed out that the IPO may not wait until 2027, but there are also reports that a postponement is being considered due to SpaceX's stock price fluctuations, so the direction is uncertain. New hardware for Codex has also been hinted at.

Google AI/DeepMind trends

The company has strengthened its price competitiveness by introducing a group of low-cost media generation models, Nano Banana 2 Lite and Gemini Omni Flash. Gemini's personalized image generation is now available to free users in the US, and Google Meet's automatic meeting minutes feature has also been expanded to Pro/Ultra plans. The biggest topic of discussion was the above-mentioned restriction on Gemini API provision to Meta, which was reported by the Financial Times and followed by multiple media outlets, attracting attention as the first case where supply-side constraints such as tight computing resources came to the fore. CNET also reported on the unauthorized download of on-device AI to Chrome, raising questions about transparency. Gemini's expansion into the automotive/smartphone control field (Renault in-vehicle integration, Android remote control) is also underway.

Trends of other major companies

In response to AI supply restrictions from Google, Meta is increasingly shifting inward to in-house development (Llama, proprietary chip Vistara). xAI has advanced the private beta deployment of Grok 4.5 and voice API integration into Vercel AI Gateway, and has also introduced real-time data integration functionality using Hosted MCP. Meanwhile, Musk himself and xAI are also facing lawsuits over the production of sexual images. Microsoft made Microsoft 365 Copilot's Service Agent GA, integrated vulnerability automatic fixing function (Copilot Autofix) into Azure DevOps, and strengthened corporate governance. DeepSeek has improved inference speed by up to 85% by releasing DSpark as open source, while also setting itself apart from China's domestic price competition by introducing API surcharges during peak hours. NVIDIA has reduced the token cost of DeepSeek v4 by up to 1/5 using Blackwell's software tuning alone.

Remarks and trends of AI celebrities

Elon Musk mentioned the possibility that Grok 4.5's performance would surpass Claude Opus and declared monthly large model releases. Meanwhile, it is also facing a battle for air pollution enforcement authority over its own AI data center. Dario Amodei's continued "unconventional" position in Washington under the Trump administration has come with costs, Fortune analyzed, and ITIF also released data disproving the myth that AI is destroying jobs. Mark Zuckerberg remains optimistic, saying he is not buying fears of job losses due to AI. Jensen Huang predicted that home AI supercomputers will become commonplace. Sam Altman continues to attract attention as the subject of Luca Guadagnino's biographical film ``Artificial'' (which was previously sold by Amazon and acquired by Neon). Yann LeCun has been mentioned more in the context of a $3 billion VC investment in the World Model space.

Technological breakthrough

DeepSeek and Peking University's DSpark (which speeds up single-user inference by 60-85% with speculative decoding) are the biggest technology topics this period. NVIDIA announced that it has reduced the token cost of the same model group to one-fifth through Blackwell software adjustments alone, demonstrating that it is possible to significantly improve efficiency simply by optimizing the software without changing the hardware. Meituan's LongCat-2.0 has been made open source with a clear statement that it will be trained using domestically produced chips. Google announced computer usage features in Gemini 3.5 Flash, making screen operations more autonomous. xAI's Grok 4.5 is a 1.5 trillion parameter model that is additionally trained using Cursor data, increasing its specialization in the coding area. OpenAI is reported to have discovered a new method that reduces inference costs by half, but the details have not been made public.

Regulation/Policy/Geopolitics

The US government's control of access to frontier models was a central issue during this period. Multiple cases have been confirmed in which Frontier Labs is coming under the de facto permission and approval of security authorities, including temporary export restrictions on Anthropic's Claude Mythos/Fable 5 and OpenAI's voluntary restrictions on new models at the request of the government. The Trump administration has issued an executive order regulating frontier AI models, and AI regulatory frameworks are being developed at the Department of Justice and each state level. Austria is calling on European countries, including Austria, to attract hosting sites such as Anthropic in response to tightened regulations in the United States. China has been the world leader in the number of API calls for large-scale AI models for nine weeks in a row, and analysis shows that the US share has plummeted from 72% to 33%, indicating that the US and China's share of model usage is in the process of reversing. Even in the legal world, the lag in generative AI governance has been repeatedly pointed out by sites such as Law.com.

Changes in the balance of power

The biggest structural change is that the scarcity of computing resources has come to the fore. Gemini API restrictions between Google and Meta indicate that the assumption that ``as long as you have data, you have unlimited computing power'' is collapsing, and competition for GPU and TPU capacity has begun among hyperscalers. Chinese companies (DeepSeek, Meituan) have clearly established a path to maintaining competitiveness in terms of performance even under hardware constraints through software optimization and open source strategies, and the fact that they have overtaken the US in terms of API usage is a full-fledged acceleration of a trend that has been going on for several months. The relationship between Frontier Labs and the government has also become polarized, from partners (California and Anthropic) to regulated targets (Mythos/Fable under export control), and the degree of government involvement has begun to be treated differently depending on the lab. xAI is a vertically integrated strategy that diverts SpaceX and Tesla's engineering resources to AI development, and is oriented toward a high-speed iteration model of monthly releases.

Points to note in the next period

First, we should pay close attention to whether Gemini supply restrictions between Google and Meta will be resolved, or whether Meta will become independent with its own model. The coming weeks will reveal whether the shortage of computing resources is a temporary spike in demand or a structural supply constraint. Second, we need to see when xAI's Grok 4.5 moves from private beta to public availability and whether Musk's declared monthly release pace actually holds up. Thirdly, whether the US government's access control to frontier models (cases like Mythos/Fable) will be extended to other labs and other models, and how the AI export control framework will be institutionalized, are issues that deserve attention as they will have a direct connection to the speed of research and development in the industry as a whole.

🟠 Anthropic / Claude114
🟢 OpenAI114
🔵 Google AI / Gemini113
🔷 Meta AI / Llama80
⚡ xAI / Grok112
🟦 Microsoft AI113
🔴 DeepSeek107
💡 AI Innovation / Breakthroughs113
⚖️ AI Regulation / Policy114
🚀 AI Products / Services112
👤 Notable People in AI596
▶️ AI Videos / Commentary80
Could Open Source AI be Banned?
Could Open Source AI be Banned?
Sentdex · 06/30 14:49 UTC
Telling my chickens that Dario won
Telling my chickens that Dario won
Matthew Berman · 06/30 19:11 UTC
Command Line Basics for Beginners - Full Course
Command Line Basics for Beginners - Full Course
freeCodeCamp.org (ML series) · 06/30 10:00 UTC
Gemma 4 in Action: Bringing Frontier AI to the Edge
Gemma 4 in Action: Bringing Frontier AI to the Edge
Google for Developers · 06/29 19:00 UTC
What’s happening in this Kotlin code?
What’s happening in this Kotlin code?
Google for Developers · 06/29 13:00 UTC
How LLMs Actually Generate Text
How LLMs Actually Generate Text
Hugging Face · 06/30 06:34 UTC
it's all bad now...
it's all bad now...
Wes Roth · 06/30 05:05 UTC
What the UK's Social Media Ban is REALLY About
What the UK's Social Media Ban is REALLY About
All-In Podcast · 06/28 20:30 UTC
2 Reasons Socialism is Rising in America
2 Reasons Socialism is Rising in America
All-In Podcast · 06/28 14:00 UTC
The Thermodynamic AI Chip · Thomas Ahle
The Thermodynamic AI Chip · Thomas Ahle
Machine Learning Street Talk · 06/28 23:38 UTC
Apple just raised prices...
Apple just raised prices...
Matthew Berman · 06/27 15:00 UTC
Run Gemma on Reachy Mini, an open source robot
Run Gemma on Reachy Mini, an open source robot
Google for Developers · 06/26 19:00 UTC
I AM SO MAD
I AM SO MAD
Matthew Berman · 06/26 20:04 UTC
GPT 5.6 is officially NOT BANNED (even worse)
GPT 5.6 is officially NOT BANNED (even worse)
Matthew Berman · 06/26 19:06 UTC
This is New Media
This is New Media
a16z · 06/26 19:58 UTC
How Hackers Think
How Hackers Think
freeCodeCamp.org (ML series) · 06/26 10:00 UTC
AI tools for human creativity
AI tools for human creativity
Google for Developers · 06/26 23:00 UTC
I can't believe this happened...
I can't believe this happened...
Matthew Berman · 06/26 21:26 UTC
What if plants could talk?
What if plants could talk?
OpenAI · 06/25 17:19 UTC
AI companies are trying to take over
AI companies are trying to take over
Matthew Berman · 06/25 19:25 UTC
Fable 5 is COMING BACK!
Fable 5 is COMING BACK!
Wes Roth · 06/25 20:07 UTC
Scaling Your AI Models with Micro-DDP – Tutorial
Scaling Your AI Models with Micro-DDP – Tutorial
freeCodeCamp.org (ML series) · 06/25 10:00 UTC
GPT 5.6 is officially BANNED
GPT 5.6 is officially BANNED
Matthew Berman · 06/26 02:24 UTC
Anthropic is coming for EVERYTHING
Anthropic is coming for EVERYTHING
Matthew Berman · 06/25 23:33 UTC
Stop writing longer prompts. Do this instead! 🧠
Stop writing longer prompts. Do this instead! 🧠
Google for Developers · 06/24 16:00 UTC
7-day Voice AI Build Challenge
7-day Voice AI Build Challenge
DeepLearningAI · 06/24 17:11 UTC
AI can run for days on its own now
AI can run for days on its own now
Matthew Berman · 06/24 20:40 UTC
GameStop CEO: 3 Ways I’d Fix eBay
GameStop CEO: 3 Ways I’d Fix eBay
All-In Podcast · 06/24 18:36 UTC
When millions of AI agents meet
When millions of AI agents meet
Google DeepMind · 06/23 15:48 UTC
Notion Workers – Full Tutorial 2026
Notion Workers – Full Tutorial 2026
freeCodeCamp.org (ML series) · 06/23 11:30 UTC
What caused this React app to crash?
What caused this React app to crash?
Google for Developers · 06/22 13:00 UTC
Can Robotics Hackathon Change Your Life?
Can Robotics Hackathon Change Your Life?
Hugging Face · 06/22 13:00 UTC
DeepSeek Just Solved AI's Billion Dollar Problem
DeepSeek Just Solved AI's Billion Dollar Problem
Two Minute Papers · 06/22 15:53 UTC
Gemma Playground: Parallel Agents in Action
Gemma Playground: Parallel Agents in Action
Google for Developers · 06/22 23:00 UTC
How One API Runs 27 ML Tasks
How One API Runs 27 ML Tasks
Hugging Face · 06/23 07:50 UTC
He won a Nobel here for AlphaFold. Then he left. - John Jumper
He won a Nobel here for AlphaFold. Then he left. - John Jumper
Machine Learning Street Talk · 06/22 22:43 UTC
Cursor just Became UNSTOPPABLE
Cursor just Became UNSTOPPABLE
Wes Roth · 06/23 08:21 UTC
NVIDIA's New Free AI - A Gift To Humanity
NVIDIA's New Free AI - A Gift To Humanity
Two Minute Papers · 06/14 15:27 UTC
Voice for AI Agents and Applications
Voice for AI Agents and Applications
DeepLearningAI · 06/17 15:00 UTC
7 INSANE loops you need to try right now
7 INSANE loops you need to try right now
Matthew Berman · 06/19 17:43 UTC
You NEED to know these vibe coding secrets
You NEED to know these vibe coding secrets
Matthew Berman · 06/18 17:44 UTC
Google's SHOCKING "POST AGI" paper...
Google's SHOCKING "POST AGI" paper...
Wes Roth · 06/18 22:11 UTC
here's REALLY WHY Fable 5 got banned
here's REALLY WHY Fable 5 got banned
Wes Roth · 06/16 22:48 UTC
What Codex Unlocks for NTT Data
What Codex Unlocks for NTT Data
OpenAI · 06/19 22:57 UTC
Plan with ChatGPT
Plan with ChatGPT
OpenAI · 06/19 21:22 UTC
← Monthly Index