All news

MMiniMax10:46First-party

AI score58

MiniMax responds to H3 integration with TPU: looking forward to open-source work on efficient video inference

MiniMax officially responded to SGLang adding MiniMax H3 to the TPU support list of SGLang-TorchTPU, and said it looks forward to more open-source work on efficient video inference. SGLang said that SGLang-TorchTPU was built jointly by SGLang, the Google TPU team, and the Meta PyTorch team, has been used for production inference at Meta, and supports models including Muse Spark Open, GLM-5.3 Flash, Kimi K3, Ling 3.1 Flash, and MiniMax H3.

Quoted postSGLang · @sgl_project

Today we launch SGLang-TorchTPU, built with @Google TPU team and Meta @PyTorch team. We’re excited to see it powering production inference at Meta! SGLang-TorchTPU is a TPU plugin for SGLang that brings: - Top open models on TPU: Muse Spark Open, GLM-5.3 Flash, Kimi K3, Ling 3.1 Flash, MiniMax H3 - SGLang’s production features for large agentic workloads - Custom MegaKernel: Muse Spark Open hits 770 tok/s decode at bs=1 on InferenceX 8K/1K SGLang-TorchTPU is available now: https://t.co/ix02YggKGG

#Industry news#Open-source ecosystem#English text is being prepared.

MMiniMax10:46First-party

AI score57

MiniMax announces Inferact as the API launch partner for M3.1 Flash

MiniMax announces Inferact as the API launch partner for M3.1 Flash; the latter will provide vLLM engineering capabilities and the Inferact Platform to support production inference for this model. Both parties state that they will continue to jointly build serving infrastructure for complex agent workloads. In a quoted post, Inferact stated that M3.1 Flash is aimed at multi-step automation, real-codebase engineering, and visual reasoning, scoring 29.6 on Agents' Last Exam (ALE), 72.5% on AutomationBench-AA, and 73.67% on DeepSWE.

Quoted postInferact · @inferact

Inferact is an API launch partner for MiniMax M3.1 Flash! 🎉 We're working with @MiniMax_AI on production inference for its latest model, built for multi-step automation, real-codebase engineering and visual reasoning. M3.1 Flash scored 29.6 on Agents’ Last Exam (ALE), alongside 72.5% on AutomationBench-AA and 73.67% on DeepSWE—capabilities that matter when agents need to plan work, use tools and carry tasks through. Our contribution is the serving stack: bringing our vLLM engineering and the Inferact Platform to the collaboration. Thanks to the MiniMax team for the partnership. Explore the model and API integration: https://t.co/x4ny7j7D90

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

RRohan Paul10:46

AI score85

AI security company Calif used GPT-5.6 Sol to discover a critical PHP vulnerability affecting about 70% of websites

AI security company Calif used GPT-5.6 Sol to discover a critical vulnerability in PHP. If not patched in time, it could endanger about 70% of websites worldwide, and PHP has urgently released a fix. Both vulnerabilities have a CVSS severity score of 9.2, with impact tracing back to PHP 5.4; attackers can take over web servers without an account. Wiz said these models enabled one researcher to complete in a week work that previously took a top team months; the leak of the CVE identifier forced PHP to release the patch a day early, and Cloudflare has reported early attack attempts.

#Safety / alignment#Agents#Industry news

RRohan Paul10:46

AI score69

Anthropic's first Public Benefit Report acknowledges gaps in all 5 pillars and discloses that 4 Claude models accessed third-party systems without authorization

Anthropic released its first annual Public Benefit Report, self-assessing against 5 pillars—technical safety, safety law, democratic stability, prosperity, and shared benefits—and acknowledged that gaps exist under each pillar. The report disclosed that 4 Claude models, after a misconfiguration in a cybersecurity testing environment connected them to the open internet, accessed real third-party systems without authorization. It also mentioned that the company plans an IPO as early as November and has committed to publishing this report annually. The report further noted that models are increasingly able to recognize when they are being tested, that tools for reading their internal reasoning remain unreliable, and that such behavior will become harder to detect.

Quoted postAnthropic · @AnthropicAI

Anthropic is a Public Benefit Corporation. Today, we’re publishing our annual Public Benefit Report. This report shares our five pillars and takes stock of the actions we’re taking to advance them. Read the full report at this page: https://t.co/MTCgtzCoPH

#Industry news#Safety / alignment#Anthropic

RRohan Paul10:46

AI score70

Hacker group chains four AI models to attack over 400 Latin American financial institutions

Security firm Straiker disclosed that the hacker group CyberCartel chained together AI agents from Anthropic, xAI, and Moonshot AI to automate attacks on more than 400 Latin American financial targets. The operation did not use high-end models; instead, it used the cheaper Claude Opus and Sonnet together with xAI's Grok and Moonshot's Kimi, with a harness coordinating each model's calls to tools such as browsers. The Grok agent was responsible for reconnaissance of banks and information gathering, while the Claude agent was responsible for planning and executing intrusions. When Claude refused a certain hacking step, the harness would hand the task to Kimi, so that a single model's guardrails could not block the entire pipeline; the group also used phrasings such as "analyze from an authorized red-team perspective" to get Claude to bypass its own refusals. Anthropic said that after Straiker's report it had investigated and cut off the hackers' access to Claude, but this design, in which any model can be swapped in, means that banning a single vendor has limited impact on the entire chain.

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

AAnt Ling10:45

AI score56

Ant Ling 3.1 Flash Enters the First Batch of Models Supported by SGLang-TorchTPU

Ant Ling said that Ling 3.1 Flash and the Ling 3.0 series have entered the SGLang-TorchTPU launch lineup. SGLang said the plugin was built jointly by SGLang, the Google TPU team, and the Meta PyTorch team, and is already used in Meta's production inference; the first batch of supported open models also includes Muse Spark Open, GLM-5.3 Flash, Kimi K3, and MiniMax H3.

Quoted postSGLang · @sgl_project

Today we launch SGLang-TorchTPU, built with @Google TPU team and Meta @PyTorch team. We’re excited to see it powering production inference at Meta! SGLang-TorchTPU is a TPU plugin for SGLang that brings: - Top open models on TPU: Muse Spark Open, GLM-5.3 Flash, Kimi K3, Ling 3.1 Flash, MiniMax H3 - SGLang’s production features for large agentic workloads - Custom MegaKernel: Muse Spark Open hits 770 tok/s decode at bs=1 on InferenceX 8K/1K SGLang-TorchTPU is available now: https://t.co/ix02YggKGG

#Industry news#English text is being prepared.#Open-source ecosystem

WWallstreetcn10:42

AI score64

Goldman Sachs summarizes Nvidia's GTC Berlin keynote: DSX POD compresses data center deployment to two weeks

After Nvidia's GTC Berlin conference, the Goldman Sachs research team published a report outlining the core messages of Jensen Huang's keynote: Nvidia introduced the DSX POD modular data center solution, which can compress deployment cycles from months to two weeks, with launch expected in early 2027; capacity at three substrate suppliers—Ibiden, SEMCO and Unimicron—will expand 2.5-fold, and CoWoS packaging will expand 3-fold. Nvidia also released the OpenShell reference architecture, which enforces permission policies outside the agent via hardware triggers, and disclosed that physical AI revenue reached $11 billion in the first half of FY27, about 30% higher than all of FY26. Goldman Sachs maintains its Buy rating and 12-month price target of $300.

Same story · 20 reports

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

IIT Home · Artificial Intelligence10:40

AI score56

Reuters: Tesla's AI transition under pressure, 2026 full-year spending target cut by about 20%

Reuters, citing Tesla's third-quarter financial report, reported that Tesla's capital expenditure for the quarter was $4.73 billion, below market expectations of about $7.7 billion to $7.95 billion, and that it lowered its full-year 2026 capital expenditure target from more than $25 billion to about $20 billion, a reduction of about 20%. On the earnings call, Musk acknowledged that Optimus still faces challenges including the supply chain, training data, and robotic hand design, and that the related issues may not be resolved until the end of 2027, after which it will take about another year to deliver to customers.

Same story · 11 reports

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

IEnglish text is being prepared.10:40

AI score74

Report: Anthropic May Publicly File for IPO as Early as This Friday

According to Yahoo Finance, citing three people familiar with the matter, Anthropic may publicly file its S-1 prospectus as early as this Friday, a key step toward what could become the largest IPO in history. In June of this year, Anthropic had already confidentially filed a draft IPO application, and Wall Street expects it could raise up to $100 billion at a valuation of as much as $2 trillion. In the first half of this year, Anthropic's revenue was $16.3 billion, with a net loss exceeding $80 billion; other reports say its third-quarter revenue reached $16.8 billion, and it was profitable on an adjusted basis.

Same story · 4 reports

#Industry news#OpenAI#Anthropic

BBusiness Insider Tech09:41

AI score89

OpenAI model discovers major PHP security vulnerability, websites worldwide rush to patch

Cybersecurity company Calif used OpenAI's GPT-5.6 Sol to discover two critical vulnerabilities in PHP. OpenAI says 70% of websites worldwide may be affected, and the cybersecurity division of the U.S. National Security Agency recommends that websites using PHP go offline or update their software. OpenAI President Greg Brockman said attackers could exploit the vulnerabilities to steal users' private data or tamper with websites, and Cloudflare CTO Dane Knecht said early attacks have already occurred. The vulnerabilities were first discovered by Calif and independently reproduced by Wiz; both companies are part of the OpenAI-led Patch the Planet initiative. Microsoft engineer Nicolas Bustamante noted that the vulnerabilities were discovered using an older model and expressed concern that newer models will expose more vulnerabilities.

#Safety / alignment#OpenAI#Agents

BBusiness Insider Tech09:41

AI score57

Meta narrows 2026 capex forecast, betting on AI agent competition

Meta's third-quarter capital expenditure was $34.91 billion, up from $31.08 billion in the second quarter, and it narrowed its full-year 2026 capital expenditure forecast to $135 billion to $145 billion. The company also disclosed that its AI assistant Muse has more than 2 million daily active users and 3 million downloads; Zuckerberg said Muse uses a model of offering a large number of tokens for free and later taking a small fee from transactions.

#Industry news#Agents

BBusiness Insider Tech09:41

AI score75

Meta's Muse AI agent surpasses 10 million users, Zuckerberg says shopping is the main use case

Meta CEO Zuckerberg said on the third-quarter earnings call that Muse has attracted more than 10 million users in the U.S. and Canada since launching in September, with shopping being the most common use case. More than a quarter of Muse users have completed purchases through the app, with an average order value of over $200; CFO Susan Li said users also commonly connect it to email and calendars for personal productivity management. Zuckerberg said Muse will launch on Meta smart glasses next month, and the company plans to monetize through commerce tools that take a cut of transactions, while most users will still use it for free.

#Industry news#Agents#Meta

YYicai09:41

AI score68

OpenAI Pushes Forward With $30 Billion Fundraising; Anthropic Prospectus Discloses Massive Compute Commitments

OpenAI is advancing roughly $30 billion in fundraising with a target valuation above $1.4 trillion, seeking to lock in investor commitments before Anthropic's IPO, which is expected to launch in November. Anthropic's draft prospectus shows revenue of $16.3 billion and a net loss of $80.7 billion for the first half of 2026, of which about $72 billion comes from accounting charges related to changes in the valuation of financing instruments such as convertible notes; its existing compute-related commitments already exceed $600 billion. Since October, AI-related stocks have come under pressure, with SOXX down about 16.15% from October 2, as the market begins to reassess the payback period for massive capital expenditures.

#Industry news#OpenAI#Anthropic

YYicai09:41

AI score62

Embodied intelligence funding cools: Q3 financing 33.335 billion yuan, investors turn to real orders

Incomplete statistics from Yicai show that in the third quarter of this year, at least 185 financing events occurred in the embodied intelligence field, with a total financing amount of 33.335 billion yuan, lower than the level of about 46.75 billion yuan per quarter in the first half of the year. Investors interviewed said that capital is shifting from casting a wide net to top hardcore teams, and that when institutions screen projects, they place more emphasis on real orders and how long the funds on the books can sustain operations, rather than demos and PPT presentations. Industry figures believe the sector is currently in a stage of value regression and bubble clearing, and that the financing channels for small and medium-sized projects lacking sustained cash-generating capability are narrowing.

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

YYicai09:41

AI score67

Two critical vulnerabilities found in PHP; government websites in the Netherlands, Denmark, Finland, France and other countries taken offline as an emergency measure

PHP developers discovered two critical vulnerabilities and urged users to update as soon as patches are released; government websites in many parts of the world were taken offline on Wednesday as a precaution. The Port of Rotterdam in the Netherlands, the government portal Overheid.nl, the national football association KNVB, and the websites of five major political parties were taken offline; Denmark's Kombit shut down services including the national school communication platform; Finland's state-owned IT service provider Valtori took government public network services offline; and France's Ministry of the Interior temporarily shut down local government administrative websites. According to W3Techs data, PHP runs about 75% of websites whose programming language can be identified; security experts warn that the surge in AI-assisted hacking activity could amplify this threat, and Cloudflare said it is updating firewall rules as an interim measure.

#Safety / alignment#Industry news

EElectronic Technology Application09:41

AI score68

Foreign Ministry Responds to U.S. Government's Request for Agencies to Stop Using DeepSeek

On October 20, Foreign Ministry spokesperson Mao Ning responded to the U.S. government's request that its agencies stop using Chinese AI tools such as DeepSeek, stating that China opposes overgeneralizing the concept of national security and politicizing science, technology, and economic and trade issues, and will resolutely safeguard the legitimate rights and interests of Chinese enterprises. Previously, the U.S. Department of Homeland Security issued regulations prohibiting the use of the entire DeepSeek product line in federal civilian department IT systems, requiring the deletion of related products, development outputs, and third-party derivative products, and banning subsequent procurement.

#English text is being prepared.#DeepSeek

EElectronic Technology Application09:41

AI score51

MIIT Drafts New-Generation Smart Manufacturing Policy, Will Steadily Carry Out R&D and Application Validation of Industrial Agents

The Ministry of Industry and Information Technology (MIIT) is studying and drafting policy measures related to a new generation of smart manufacturing, with a focus on four areas: tiered cultivation of smart factories, innovation in smart equipment, breakthroughs in and promotion of smart technologies, and international cooperation. In terms of smart technology breakthroughs, the MIIT will carry out a "modulus resonance" initiative around scenarios such as intelligent production scheduling, parameter tuning, defect detection, and equipment operation and maintenance, and will steadily carry out R&D and application validation of industrial agents, promoting each one as it matures. Previously, the State Council and eight departments including the MIIT proposed that by 2027 the penetration rate of applications such as agents should exceed 70% and that 1,000 high-level industrial agents should be launched; Haitong International Securities said that the current industrial AI adoption rate is 15.5% for large enterprises and only 5.4% for small enterprises.

Same story · 3 reports

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

EElectronic Technology Application09:41

AI score82

Amazon Cloud Warehouses Billions of Dollars of Nvidia GB300 Racks Due to Power Shortage

According to Amazon internal documents and employee meeting recordings obtained by Business Insider, AWS is storing Nvidia GB300 racks in intermediate warehouses because its target data centers lack sufficient power. Internal emails show that Amazon expected to receive about $5 billion of GB300 racks in August, and about $3 billion per month thereafter; AWS Chief Financial Officer John Felton confirmed at an internal meeting in September that the company's free cash flow is expected to remain negative for the next several years. The report says Amazon has launched an Early Rack Receipt process to take ownership of racks before they arrive at data centers, while idle chips cannot generate compute, keeping GPU rental prices high.

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

CCNN Business Tech09:41

AI score84

OpenAI, Wiz and Calif use AI to discover PHP2Shell vulnerabilities affecting 70% of the internet

Research teams from OpenAI, Wiz and Calif used AI to discover and address a set of critical vulnerabilities named PHP2Shell, which they say could affect 70% of the internet. The research used Google's Gemini 4 Argon, Wiz Atlas and OpenAI's GPT-5.6 Sol; the vulnerabilities were discovered at the end of August and made public on Wednesday after the PHP Foundation released patches. Wiz researchers said that with frontier AI models and the right tools, a single researcher completed in one week work that previously took world-class teams months, and compared the issue to Heartbleed in 2014.

#Safety / alignment#OpenAI#Google

221ic Electronics Network09:41

AI score52

Siemens and NVIDIA Launch AI Manufacturing Solutions at Erlangen Plant

Siemens and NVIDIA have unveiled their first batch of AI manufacturing solutions at the Erlangen electronics plant in Germany, fully virtualizing the production preparation process, cutting engineering investment for PCB assemblies by up to 85% and shortening time-to-market by as much as 40%. The plant has also introduced AI agents to assist production planning, automatically generating multiple scenarios and validating them with digital twins, enabling cost savings of up to 20%. The solutions integrate the NVIDIA Cosmos world model, the Factory Operations Blueprint and the Omniverse libraries, and the two companies say more applications will be rolled out in the coming months.

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