24-hour AI news: EU's "Dual Strategy" officially released, Google releases browser-level AI model

In the past 24 hours, overseas, the European Union officially released the two major strategies of "Applied AI" and "AI in Science", Google launched the Gemini 2.5 "Computer Use" model that can be directly operated in the browser, the US Patent and Trademark Office launched an AI pre-examination search pilot, and OpenAI released the October threat intelligence update; domestically, SenseTime and Baidu Auto have joined forces to create an automotive AI "super brain", Alibaba Tongyi Qianwen is preparing to establish a robotics team, energy companies have released industry large models, and new embodied intelligent products have been launched, forming a relatively balanced new progress matrix at home and abroad.

1. The EU officially released two major strategies: "Applied AI" and "AI in Science"

1. On October 8, the European Commission officially announced two major strategies, with supporting financial support of approximately 1 billion euros (approximately 1.1 billion US dollars), focusing on key areas such as medical care, energy, manufacturing, automobiles, defense and scientific research.
2. Initiatives include building advanced AI-enabled medical screening centers, promoting agent projects in manufacturing and climate science, and coordinating funding through channels such as Horizon Europe and Digital Europe.
3. Compared with the previous interim notification of "imminent implementation", this measure has entered the stage of "formal implementation", marking the shift of Europe's "artificial intelligence first" thinking from regulations to applications.

Google releases Gemini 2.5 "Computer Use" model: allows users to operate their browsers like humans

1. The new model released on October 8 can complete in-browser operations such as "view the interface - click buttons - fill in forms - drag and drop" without a dedicated API, and is adapted to complex human-computer interfaces.
2. Officials claim that it outperforms similar solutions on multiple web and mobile benchmarks, and has built-in 13 types of safe action constraints and protection strategies, which are open to developers in AI Studio and Vertex AI.
3. Identify the key capabilities of "Agentic AI" (intelligent entities with autonomous task execution capabilities) to reduce costs and increase efficiency in scenarios such as automated UI testing and long-link web page process processing.

3. USPTO Launches AI Pre-examination Search Pilot Program: Accelerating Patent Preliminary Examination Efficiency

1. On October 8, the United States Patent and Trademark Office launched the "AI Pre-examination Search" pilot program to search and compare invention patents in the pre-examination stage and optimize the examiner workflow.
2. The goal is to improve the coverage and speed of recognition of similar prior art while ensuring interpretability and reviewability.
3. If the pilot program is effective, it may be expanded to include applications in more technical fields, which will indirectly benefit the efficiency of intellectual property rights confirmation in high-iteration industries such as AI and semiconductors.

4. OpenAI Releases October Threat Intelligence Update: AI More Like an Accelerator Than a New Weapon

1. The external update from October 7-8 showed that more than 40 networks that violated the usage policy had been dealt with in the past quarter, mainly involving the acceleration of old methods such as phishing, fraud and cognitive operations.
2. The report states that no evidence was found that the model brought about a "new leap in offensive and defensive capabilities." Instead, it more commonly "improved efficiency and speed" in areas such as writing, translation, and code snippet search.
3. In some cases, overseas organizations attempted to use chat models to monitor proposals and write phishing materials. The platform has banned the relevant accounts and notified partners.

5. SenseTime and PATEO Automotive Jointly Build an Automotive AI "Super Brain"

1. On October 8, the two parties officially announced their strategic cooperation, deeply integrating large models in the intelligent cockpit and jointly planning cutting-edge directions such as "cabin-driver integration" and humanoid robots.
2. Based on the combination of the "Qinggan Big Model" and the daily new big model, create a platform-level in-vehicle intelligent body, covering entertainment, local services, communication collaboration, travel navigation and other capabilities.
3. Engineering implementation of end-cloud collaboration and inference acceleration will be implemented, which is regarded as a milestone for car companies to move from "functional stacking" to "scenario-based intelligent services."

6. Alibaba Tongyi Qianwen is establishing a robotics team, extending the AI battlefield into the physical world.

1. According to reports, Tongyi Qianwen has formed an internal team to enter the field of robotics and promote the research and development path of "model × embodied intelligence".
2. In the short term, focus on deterministic scenarios such as factories and commercial services, combining large language models with multimodal capabilities of control and perception.
3. This trend is consistent with the domestic "artificial intelligence + manufacturing" policy and industry rhythm, and is expected to form a long-term moat integrating software and hardware.

7. Energy scenarios add another industry model: Langxin's "Jiugong" AI energy model released

1. On October 8, Longxin Group launched the "Nine Powers" model for energy systems, focusing on the dual engines of time series prediction and AI intelligent agents.
2. Covering nine major functions including load/generation forecasting, scheduling optimization, transaction matching and risk warning, and green electricity consumption.
3. Cooperate with its existing businesses such as the aggregated charging platform, new energy trading and photovoltaic cloud platform to provide end-to-end closed-loop capabilities for the implementation of "AI+Energy".

8. Nankai University-related companies released new embodied intelligent products and unveiled a three-layer technical architecture

1. On October 8, a local company in Tianjin released a new embodied intelligent product based on "large model access - intelligent body scheduling - memory module".
2. It has been adapted to multiple mainstream basic models and tool chains, emphasizing remote control and low latency.
3. Reflect the technology transformation path of university-industry-enterprise linkage, and inject new samples into the local "AI+Robot" ecology.

9. Anthropic announced it will establish its first India office in Bengaluru (opening in 2026)

1. On October 8, it was announced that Bangalore would be the site of its launch, making India one of its key markets in the Asia-Pacific region.
2. Claude's usage among developers and enterprises in India has grown significantly, and officials said they will provide supporting local cooperation and ecological investment.
3. This layout is consistent with the expansion trend of "localization + regional compliance" of global AI companies.

10. Financial and Regulatory Trends: Rebalancing AI Valuation and Risk

1. The Bank of England’s Financial Policy Committee warned on October 8 that valuations of AI-related assets are high, and if expectations are revised or infrastructure is restricted, it may trigger a market pullback.
2. On the same day, many security and industry media also focused on the "real-world efficiency enhancement - not subversion" attribute of AI in the fields of social engineering and compliance.
3. The interplay of hot and cold on the industry and capital sides suggests that companies need to pay attention to security, computing power, and compliance costs when promoting the implementation of AI.

Frequently Asked Questions (Q&A)

Q: What is the difference between the EU’s “dual strategy” this time and the previous announcement of “imminent implementation”?

A: This is the "official release" on October 8, which clarifies the funding scale (approximately 1 billion euros), implementation areas and demonstration projects, and enters the implementation stage compared to the previous announcement.

Q: What is the strength of Gemini 2.5 "Computer Use"?

A: The biggest difference is "being able to read the interface and click". It can complete complex web processes without API. It is suitable for e-commerce forms, process automation and UI testing, etc., and has action-level safety guardrails.

Q: What impact does the USPTO's AI pre-examination search pilot have on companies' patent applications?

A: It is expected to shorten the time for early search and comparison, improve the comprehensiveness and verifiability of similar technology identification, and help high-iteration industries (AI/chips/software) enter the substantive examination stage faster.

Q: Does OpenAI’s threat intelligence update mean that “AI has become a new weapon for hackers”?

A: The report leans towards "efficiency improvement rather than disruption." Common examples include using AI to write emails, translate, and draft scripts. This model hasn't been proven to bring about a new leap in attack capabilities, and platforms are continuing to block and trace the source.