All news

Hheise online22:46

AI score55

US government refuses to disclose safety review rules for frontier AI models

In responding to Protect Democracy's FOIA lawsuit, the US government released only some documents and refused to disclose the full content of the voluntary safety review framework for frontier models, accompanying documentation, and the list of companies granted early access to models, citing national security and trade secrets. The review is based on Executive Order 14409 signed by Trump in June 2026, which allows developers such as Anthropic and OpenAI to voluntarily open their models to the government up to 30 days before release, but which models qualify as covered frontier models and what testing and safety standards are used remain undisclosed. Protect Democracy criticized the executive branch for being able to decide which AI products are released and which customers gain access without public standards, and for the inability of external institutions to reproduce the tests.

#English text is being prepared.#Safety / alignment#OpenAI

Hheise online22:46

AI score62

Google plans to support a unified AI training policy token in robots.txt starting in 2027

Google plans to support a unified policy token in robots.txt (__usage_ai-training) starting in 2027, allowing website operators to use a single entry to express, to multiple AI vendors participating in the mechanism, whether content may be used for model training, without having to maintain vendor-specific rules such as GPTBot, ClaudeBot, and Google-Extended one by one. The proposal is based on prior work by the IETF AIPREF working group and follows existing User-Agent syntax; parsers that do not recognize the token will ignore it as an unknown User-Agent under RFC 9309. Google says that if a site prohibits AI training via this token without separately specifying Google-Extended, it will automatically be treated as having blocked Google-Extended; and explicitly allowing Google-Extended is also a prerequisite for a site's content to appear in Grounding answers in the Gemini app.

Same story · 3 reports

#English text is being prepared.#English text is being prepared.#Google

SServeTheHome22:42

AI score62

Qualcomm Completes Preliminary Validation of HBC Stacked Memory, Dragonfly AI250 to Begin Sampling in First Half of Next Year

Qualcomm announced that it has completed preliminary validation of its High Bandwidth Compute (HBC) stacked memory architecture, which stacks multiple layers of DRAM on a small compute die for use in its Dragonfly data center AI accelerators. The testing used early samples with 2 layers of DRAM, while production products will have 4 layers; Qualcomm claims that its energy efficiency per unit of bandwidth per watt exceeds the HBM4 specification by 5 times, and that measured bandwidth is 18 times that of external LPDDR5X, but it did not disclose specific bandwidth figures. The first Dragonfly AI250 integrating HBC will begin sampling in the first half of next year, and Qualcomm also plans to use this technology in Snapdragon consumer products.

#Industry news#English text is being prepared.#English text is being prepared.

99to5Mac22:42

AI score66

Wired Investigation: Over 400 AI Apps with Hidden Pornographic and Violent Deepfake Features Passed App Store Review

A Wired investigation found that a network suspected of being operated by Chinese entities includes more than 400 domains and apps. These apps passed review by presenting everyday features such as language teaching and AI fitness coaching, then loaded interfaces capable of generating pornographic and violent deepfake content under specific conditions. They drew traffic through ad placements on Meta's platforms, and some apps had been live for a year. After Wired provided details, Apple removed 19 apps, saying the developers had swapped the source content websites after listing; Meta said it had taken action against more than 10,000 ad accounts associated with these app domains.

Same story · 2 reports

#Safety / alignment#Image generation#Video

SSina Tech22:29

AI score56

OpenAI pledges that its self-built U.S. data centers can cut power load during extreme weather

OpenAI announced the launch of the "Grid Guardian Plan," under which data center campuses it directly builds in the United States will proactively reduce electricity consumption when extreme weather or other emergencies put stress on the grid. The plan extends to the entire country a commitment already made for a planned campus in Effingham County, Georgia; OpenAI pledges to cut up to 1 gigawatt of power load when the grid needs relief, roughly equivalent to the electricity consumption of 700,000 households. John McCarrick, OpenAI's head of energy policy, said this reflects a shift in industry thinking and technology, as well as adjustments made after incorporating community feedback.

Same story · 5 reports

#Industry news#OpenAI#English text is being prepared.

CcnBeta22:27

AI score58

OpenAI launches Grid Protectors Program, data centers can cut up to 1 gigawatt of load in extreme weather

OpenAI announced the "Grid Protectors Program," extending a commitment for a planned data center campus in Effingham County, Georgia, to the entire United States, with the ability to cut up to 1 gigawatt of electricity load when the grid needs relief, roughly equivalent to the electricity consumption of 700,000 households. OpenAI said it will negotiate with power companies to determine the scale of reducible load, response speed and duration, and details for each campus will be announced separately once finalized. John McCarrick, OpenAI's head of energy policy, said this reflects a shift in industry thinking and technology, and noted that a year ago some large cloud providers still insisted on the "five nines" reliability standard of 99.999% continuous uptime.

Same story · 5 reports

#Industry news#OpenAI#English text is being prepared.

SShanghai Observer22:22

AI score55

AgiBot's 6 Genie G2 Robots Work 24-Hour Live-Streamed Shifts at a Convenience Store

AgiBot launched a 24-hour livestream at an operating convenience store in Xuhui District, where 6 Genie G2 robots took on tasks including greeting customers, making coffee, serving prepared food, picking and restocking goods, and cleaning tables, working in 4 shifts under real-time supervision by the entire internet. As of 2:30 p.m. yesterday, the 6 robots had been on duty for 4.5 hours and completed a total of 337 tasks, including greeting customers 116 times, delivering and retrieving 84 items, and making coffee 30 times. AgiBot previously released the GO-3 foundation model and the embodied intelligence agent "Genie Master," and together with Tsinghua University and Infinigence AI proposed a collective experience learning paradigm, validated in a cluster of 96 robots, where the success rate rose from 86.7% to 98.3% as the number of robots participating in training increased from 16 to 96.

Same story · 6 reports

#Industry news#Embodied AI#Agents

SShao Meng21:41

AI score62

DeepSeek Search and Fetch End Free Tier, Billed at $0.75 per Thousand Calls

The two web tool APIs, DeepSeek Search and Fetch, end their free tier starting today and switch to billing by the number of successful requests, at the same unit price of $0.75 per thousand calls. Billing only counts requests that return HTTP 200 and where Search returns non-empty results or Fetch actually retrieves page content; failures, empty searches, and timeouts are not billed, and the cost is independent of the number of returned items, tokens, and whether need_content is enabled. Rate limits are calculated per account: Search is 100 QPS and Fetch is 25 QPS, with Fetch also having a per-domain QPS cap; server-side tool calls initiated via DeepSeek Harness, Claude Code, and Codex are subject to the same rules.

#Industry news#DeepSeek#English text is being prepared.

RRohan Paul21:40

AI score60

TrendForce: NVIDIA Rubin Transition Begins, High-End GPU Lead Times Extended by 5 Weeks to 45 Weeks

TrendForce data shows that with the start of the NVIDIA Rubin transition, high-end GPU lead times have been extended by 5 weeks to 45 weeks. Hyperscale cloud providers may get Rubin first, while enterprise buyers continue to use existing GPU platforms. On the memory side, arrivals of 96GB and 128GB RDIMMs have accelerated, while 32GB and 64GB modules remain severely tight; DRAM lead times have shortened by 1 week to 19 weeks, still more than twice normal levels. Because servers cannot ship until the slowest component arrives, Rubin's capacity ramp depends on substrates and capacitors whose lead times exceed balanced levels by 3x.

#Industry news#English text is being prepared.#English text is being prepared.

CCailian Press21:28

AI score57

Communities Across the U.S. Oppose Data Center Construction, Adding Uncertainty to Midterm Elections

Twenty U.S. communities will vote on pausing or banning local data center construction, 13 of which are located in Republican-dominated districts. A POLITICO poll shows that 36% of Republican voters and 65% of Democratic voters oppose building data centers within 3 miles of their homes. Oracle's $165 billion Jupiter project in New Mexico faces delays, and an executive order in New York State pausing environmental permit approvals for large facilities has already delayed nearly 1GW of data center construction.

#Industry news#English text is being prepared.#OpenAI

CcnBeta21:26

AI score55

South Korea's exports approach $1 trillion, with the AI memory chip boom as the biggest driver

South Korea's exports are approaching the $1 trillion mark, with semiconductors as the biggest driver. In October, semiconductor exports reached $56.44 billion, up 259.4% year-on-year, accounting for 49% of total exports that month. Amid the AI infrastructure construction boom, demand from data centers for advanced memory products has risen, and Samsung Electronics and SK Hynix have continued to benefit; over the same period, computer equipment exports grew 496.6% year-on-year. However, the semiconductor boom has not fully transmitted to South Korea's domestic economy: third-quarter retail sales fell 0.5% year-on-year, and SMEs face weak sales and operating pressure.

#Industry news#English text is being prepared.

LLeiphone21:07

AI score60

Alibaba's Qoder Wins World Internet Conference Leading Science and Technology Award, Has Served Over 8 Million Users

Alibaba's agent platform Qoder won the Leading Science and Technology Award at the 2026 World Internet Conference. It is one of 20 selected achievements and the only general-purpose Agent product in China to receive this award. Qoder has been released globally since August 2025, with a product line covering the desktop version, AI-native IDE, CLI, JetBrains plugin, QoderWake, and cloud-based Cloud Agents. It has now served over 8 million users and more than 100,000 enterprises worldwide. The original article also mentions that the Gartner AI Coding report placed it in the Challengers quadrant, and that in IDC's 2025 China AI coding market share report, Qoder ranked first with 47.6%.

#Industry news#Agents#English text is being prepared.

LLeiphone21:07

AI score53

Alipay AHA Multi-Terminal Agent Interconnection Solution Wins World Internet Conference "New Glory" Scenario Award

On November 2, the 2026 World Internet Conference Wuzhen Summit opened, and Alipay's AHA (Agent Hub Access) multi-terminal agent interconnection solution won the "New Glory" Scenario Award. AHA uses the Alipay App as the anchor for accounts and capabilities, and builds a three-layer interconnection protocol system based on the open-source A2A, MCP, and A2UI protocols, exporting Alipay's government services, transportation, healthcare, life services, and other capabilities to terminals such as phones, car infotainment systems, smart glasses, and large-model apps. The solution was officially released to the public in August this year. Currently, Abao's service capabilities have been exported to 6 types of terminals and deployed in more than a thousand service scenarios, and it has built an ecosystem with more than 20 enterprises, achieving cross-terminal collaboration with 5 major smartphone manufacturers and covering 16 automobile brands on the car side.

Same story · 2 reports

#Industry news#Agents#English text is being prepared.

SSina Tech20:27

AI score71

Marvell Announces All-Stock Acquisition of Lumentum, Implied Enterprise Value of $115 Billion

Marvell Technology and Lumentum have signed a definitive agreement under which Marvell will acquire Lumentum in an all-stock transaction, with each Lumentum share exchanged for 4.500 shares of Marvell common stock, implying an enterprise value of $115 billion and a per-share price of approximately $1,132, a premium of about 23.9% over the closing price on October 28, 2026. Upon closing, former Marvell shareholders will hold approximately 67% of the combined company and Lumentum shareholders approximately 33%; the combined company's total addressable market is expected to expand to $550 billion, with annualized cost synergies expected to exceed $1 billion. The transaction is expected to close in the first half of 2028, subject to shareholder votes by both companies and regulatory approvals in various countries.

Same story · 9 reports

#Industry news#English text is being prepared.

Tt3n20:17

AI score65

Anthropic Bans Claude from Replicating Existing Games, Affecting Legitimate Projects Too

Anthropic added a clause to the system prompts of Claude Code and Claude Opus 5.5 prohibiting the replication of specific existing games, even if developers change the names, and the company confirmed that this move is to avoid intellectual property infringement. The original text states that the restriction also affects legitimate projects: Claude will refuse requests such as analyzing game mechanics or patching game code, and will even refuse to analyze Windows crashes or clean up junk software on smart TVs; Anthropic employee Addy Osmani said the rules will be adjusted to find a balance.

Same story · 15 reports

#English text is being prepared.#Anthropic#Agents

SSolidot19:52

AI score58

Proton pushes engineers to shift to AI, CEO says security concerns are loser mentality

Proton, known for its privacy services, is going all out to push engineers to adopt AI. CEO Andy Yen reprimanded employees who raised security concerns in an internal Slack channel, calling it a loser's mentality. Proton also planned to re-interview employees and require them to explain how they use AI; employees who do not use AI sufficiently may face dismissal, and the re-interviews were canceled after public backlash. Proton's published AI manifesto states that engineers' task is no longer to write code, but to help LLMs write code faster and better.

#Industry news#English text is being prepared.#English text is being prepared.

AAnalytics India Magazine19:45

AI score62

OpenAI signs four agreements with Abu Dhabi covering government, energy and investment

OpenAI signed four agreements with multiple Abu Dhabi institutions to deploy its AI models in the government, energy, investment and education sectors, following a meeting between CEO Sam Altman and Abu Dhabi Crown Prince Sheikh Khaled bin Mohamed bin Zayed Al Nahyan. On energy, ADNOC, G42 subsidiary Inception42 and OpenAI are collaborating to run GPT models on inference infrastructure within the UAE, with ADNOC saying it became the first company in the energy industry to deploy OpenAI's frontier AI technology. On government, the Abu Dhabi government and OpenAI launched the Machine-Readable Government initiative to convert some policies and regulations into machine-readable formats, with the first pilot already in use at the Abu Dhabi Agriculture and Food Safety Authority for food import assessments, while officials retain final decision-making authority.

Same story · 6 reports

#Industry news#OpenAI#English text is being prepared.

TThe Guardian Technology19:42

AI score66

Guardian investigation: NHS says 29 hospitals benefit from Palantir platform, but these hospitals say they do not use it or rarely use it

Through more than 100 freedom of information requests, the Guardian found that the 29 hospital trusts publicly listed by NHS England as "realizing benefits" from Palantir's £330 million Federated Data Platform (FDP) said the system was not deployed or that they did not use any of its applications. NHS England said nearly 140 trusts have reported benefiting from the platform's tools, and explained that the benefit list includes organizations that only use the tools for back-end data analysis and submission; the UK Statistics Authority said it will incorporate the relevant data claims into its case handling. Chi Onwurah, chair of the UK Parliament's Science and Technology Committee, and the Health Select Committee called on the government to activate the 2027 contract break clause and to openly tender for a UK-based supplier.

#Industry news#English text is being prepared.

NNumerama19:40

AI score58

Anthropic requires Claude Code to refuse to replicate existing games

When using Opus 5.5 in Claude Code, Anthropic added two instructions requiring it to refuse to replicate existing games, even if users rename them or claim they are original. Addy Osmani, a member of the company's technical team, confirmed this on October 30 and said it will still support original games, games of the same genre, and projects using their own assets. Some users reported that legitimate tasks were refused, but some of those cases may stem from Claude's cybersecurity protections rather than this game rule.

Same story · 15 reports

#Product updates#English text is being prepared.#Anthropic

LLeiphone18:45

AI score53

Details of DeepSeek's second funding round of 105.5 billion yuan revealed, with vivo, Meituan, Xiaohongshu and others among investors

DeepSeek's second funding round has raised 105.5 billion yuan (approximately US$15.8 billion), attracting nearly 30 investors, with state-owned capital contributing about 27%, a sharp increase from 2% in the first round. Tencent Holdings and CATL remain the two major investors; Tencent contributed 10 billion yuan in each of the two rounds, while CATL and its chairman Zeng Yuqun contributed 5 billion yuan and 10 billion yuan in the two rounds respectively. Other industrial investors also include ChangXin Memory Technologies, SMIC, vivo, Meituan, Geely Automobile, Midea Group, Luxshare Precision, Ctrip and Xiaohongshu, among others, with some companies investing between 1 billion and 3 billion yuan each.

Same story · 5 reports

#Industry news#Agents#DeepSeek