2026-10-14
AI Briefings Daily Report.

Deezer to launch toggle for filtering AI-generated music ↗

Deezer is developing a toggle that allows users to filter out AI-generated songs, albums, and related artists from search results; the feature is expected to launch before the end of this year. Deezer stated that a survey last year showed about 45% of users wanted to remove AI music, while in July this year, AI-generated content accounted for over 50% of daily uploads on its platform, averaging about 90,000 tracks per day.

Zenzap raises $44 million to replace workplace WhatsApp with an AI chat app ↗

Zenzap completed a $44 million funding round, with investors including Bessemer, Lightspeed, and The Garage, to build an enterprise-focused private AI messaging application.

Floqer plans to open-source its AI agent-based customer knowledge database ↗

Floqer announced plans to open-source its proprietary customer knowledge database, targeting an early 2027 release pending security review. The platform uses AI agents to automatically research potential customers across more than 100 third-party data sources and continuously update CRM records; customers report prospecting time reduced by up to 70%. The company stated that ARR has exceeded $1.2 million, with overall revenue growing approximately tenfold over the past year, and currently serving more than 1,000 clients.

MIT team doubles accuracy of two-to-six-week weather forecasts using machine learning to correct errors ↗

A research team from MIT published a paper in Nature, proposing a machine learning post-processing layer that learns errors from existing forecast models using historical data and corrects their outputs, doubling forecast accuracy for the two-to-six-week range. The model ranked first in the ECMWF seasonal forecasting competition, becoming the first to consistently exceed climatology averages within this time frame, and significantly improving predictions for extreme weather events such as floods and cold waves; the model is open-source, low-cost, and can be attached to any forecast model, with the Kenya Meteorological Department already in contact to use it for improving monsoon season forecasts.

Intel Launches Inference Microservice for Fourth to Sixth Generation Xeon Processors: Automatically Detects Hardware and Configures vLLM Parameters ↗

This containerized microservice automatically detects hardware and applies pre-validated configuration parameters, eliminating the need for manual tuning. It will later extend to Intel GPU platforms, with the interface remaining unchanged.

Man Sues for Million Yuan After Posting AI Rumors About AITO ↗

According to Deep Blue Finance, Seres Automobile Co., Ltd. disclosed that it has filed a civil lawsuit against a short-video user on Douyin. The user used AI tools to synthesize false footage showing the body of an AITO vehicle collapsing while driving and posted it online to smear the vehicle's safety performance. Seres demands that the defendant delete the infringing statements, issue a written public apology, pay 1 million yuan in damages for infringement, and bear the relevant litigation costs of the case.

Sakana AI launches memory visualization feature My Fish for Sakana Chat ↗

Sakana AI launches memory visualization feature My Fish for Sakana Chat, allowing users to see what the chatbot has remembered, when memories were used, and how they are connected. The feature supports three views: overview, connection graph, and historical timeline. Memories unused for over 30 days will fade, and users can turn off the memory feature at any time in settings.

TechCrunch Disrupt 2026 Day 2: Nvidia, Waabi, and others discuss AI hardware and robotics ↗

The second day of TechCrunch Disrupt 2026 was held at Moscone West in San Francisco, focusing on AI, robotics, AI hardware, public safety, and venture capital.

a16z: How Long-Horizon Agents Win the Relay Race for High-Value Customer Journeys ↗

a16z partner David Haber wrote that long-horizon agents can now handle complete customer journeys spanning dozens of steps over weeks or months, such as insurance claims, mortgages, and enterprise sales.

Character.ai bankruptcy becomes the first domino in the collapse of the AI bubble ↗

Character.ai's bankruptcy has been called the first domino in the collapse of the AI bubble by 404 Media. This item was included in the paid member newsletter on October 14, which also mentioned online trends such as Tesla locking police out of cars.

Jensen Huang announces $2 billion investment in Europe and donates $250 million in computing power to a UK safety lab ↗

Nvidia CEO Jensen Huang announced a new $2 billion investment in Europe in Paris, to be deployed within the following week; Nvidia spent approximately $8 billion in Europe last year. He also disclosed new stakes in French startups Doctolib, Genesis AI, Pigment, and Entalpic, and promoted open-weight models to help Europe close the AI gap.

Reports indicate Nothing will launch two smart glasses: an audio version and a camera version, with the latter supporting ChatGPT. ↗

Sources say Nothing will launch two smart glasses: one supports music playback, calls, and voice assistant interaction, equipped with photochromic lenses; the other features a camera and supports ChatGPT and translation. It is reported that AI tasks are handled by the phone and cloud, with no AR display included. ##Nothing Smart Glasses##

Available to both new and existing users: Nubia NaviX Ultra mobile phone adds purchase benefits, including a complimentary six-month Doubao Standard Edition membership. ↗

During the first evolution live stream for the Nubia NaviX Ultra tonight, Nubia Mobile officially announced additional purchase benefits for the NaviX Ultra, available to both new and existing users: Users who post their purchases during the Double 11 festival will receive an additional six-month membership benefit for the Doubao Standard Edition. Users who purchase in the first month directly enjoy this benefit. The official value of this benefit is stated as 408 yuan.

Artemis Security launches Orion-1, a security-specific model, claiming it surpasses Claude Opus 5.5 and GPT-6 Sol on its own benchmarks ↗

U.S. startup Artemis Security released Orion-1, a security-specific model capable of autonomously investigating and responding to corporate cyberattacks, which outperformed Claude Opus 5.5 and GPT-6 Sol on the company's proprietary benchmarks.

Anthropic warns that small AI models may achieve self-replication via IoT devices ↗

Anthropic issued a warning that progress in small AI models could spawn malware capable of autonomously self-replicating on IoT devices such as smart refrigerators and robot vacuums. By next year, small models may replicate across various IoT devices, eventually forming computer worms that spread without human interaction. Its blog notes that such replicants have no centralized shutdown point, are difficult to stop, and can handle unfamiliar devices independently, but recent company experiments show that achieving this kind of AI malware replication still faces significant challenges.

OpenAI engineer arrested for allegedly using wife's chatbot account to assist in bugging executive meeting rooms ↗

The U.S. Department of Justice announced that Abdulrahman Alfozan, a Saudi national engineer at OpenAI, was arrested on October 9 for allegedly hiding audio recording equipment in the company's executive meeting rooms. The indictment states that he obtained prompts for building the recorder, Wi-Fi connectivity, and SD card encryption through a chatbot account registered under his wife's name; the device was hidden in a TV USB port and wrapped in toilet paper. If convicted, he faces up to 5 years in prison and a fine of $250,000.

5,499 → 3,823 yuan: DJI AI ROMO 2 Extra-Wide Sweep-and-Mop Robot Official Price Drop for 11.11 ↗

DJI released the ROMO 2 series in May this year, including the ROMO P2 and ROMO A2.

Zhu Jun, founder of Shengshu Technology, wins the 2026 CCF Youth Science Award (CCF-ACM Artificial Intelligence Award) ↗

Zhu Jun, founder of Shengshu Technology and professor at Tsinghua University, has been selected for the 2026 CCF Youth Science Award, receiving the CCF-ACM Artificial Intelligence Award jointly chosen by the China Computer Federation and ACM.

Modem migrates to modem.ai and releases Modem 2.0 semantic observability platform ↗

Starting this week, Modem is using a new domain, modem.ai; old modem.dev links will automatically redirect. Simultaneously, Modem 2.0 has been released—a rebuild of its semantic observability platform that connects user channels, identifies and categorizes user concerns, and tracks team handling via PR and ticket monitoring, with improvements in both accuracy and end-to-end performance. Modem states its user base has expanded from engineers to founders, PMs, customer success teams, and CTOs, among other roles; the new domain aims to move away from being perceived as a pure developer tool.

Apple iOS 27.2 Beta 4 system Siri AI adds support for French, Japanese, Spanish, and more ↗

Apple today pushed the iOS 27.2 Beta 4 developer preview, adding support for French, Japanese, Korean, Portuguese, and Spanish to Siri AI, bringing the total to six languages including English. The AI assistant can understand personal context, execute app actions, perceive the screen, and access world knowledge, and now features a dedicated app.

The career prospects of software programmers over the next 10–15 years in the AI era ↗

Software development will not disappear, but programmers whose primary value lies in 'writing code' will face ongoing reassessment of their professional worth. The future may demand more software, but not necessarily a proportional increase in programmers.

Lao Gan Ma quietly earned 5.4 billion yuan using AI. Is your AI still just writing copy? ↗

--- | --- | --- | --- | | 1 | Form submission per station + accounting entry | Station managers spend one hour daily filling out Excel sheets, and accountants spend another hour entering data into financial software. Five dedicated accountants handle this task. | The annual salary cost for these five accountants amounts to several hundred thousand yuan. | Current spreadsheet

BYD x Volcengine: Doubao Large Model Atomic Capabilities Injected into 'Didi Xia' ↗

Scimplify raises $25 million, using AI platform Scinode to challenge China's dominance in key materials ↗

Indian startup Scimplify has completed a $25 million all-equity strategic funding round led by Hitachi Ventures, reaching a post-money valuation of $250 million. Its AI platform, Scinode, uses specialized agents to coordinate R&D, manufacturing, and quality control, reducing order fulfillment from demand to delivery to just two months for some orders, compared to the usual 12 to 18 months.

Honor MagicOS 11 October Upgrade Features Announced: AI Gesture Control Update; IT Home News / Community Homepage and Topic Page Floating Buttons Now Support Drag-and-Drop ↗

The Honor MagicOS official account today announced the highlights of the October experience upgrade for MagicOS.

Dell CEO: AI requires safeguards and controls; government should play a role. ↗

Michael Dell uses fire as a metaphor, stating technology can warm homes or burn down villages, emphasizing that AI models must have safety measures and clear permission boundaries, and that governments should also participate in regulation. #DellCEOOnAISafety# This statement comes amid industry debate over the potential risks of self-improving AI.

Study finds OpenAI and Google's new image models vulnerable to RISE method bypassing safety filters to generate nude content. ↗

Security firm White Circle proposes the RISE attack method, using off-the-shelf tools and API calls to bypass safety filters in DALL·E 3, Google Nano Banana 2, and GPT-Image-2, with approximately 13% of attempts generating explicit content.

BYD x Volcengine: Doubao Large Model Atomic Capabilities Injected into 'Didi Xia' ↗

Volcengine provides core atomic capabilities such as Doubao Large Model text-to-speech (TTS) for BYD's super intelligent agent 'Didi Xia', currently covering five major brand models: Yangwang, Denza, Fang Cheng Bao, Dynasty, and Ocean. The Doubao speech synthesis model can predict emotion, tone, and rhythm based on context, while the online Q&A agent provides real-time knowledge Q&A and multi-turn dialogue support. Both parties are also refining innovative applications of Doubao's real-time voice dialogue model and text-to-image capabilities.

Andrew Stellman analyzes why AI writes Yoda-style inverted sentences ↗

Andrew Stellman proposes a theory and uses experiments to explain the causes of Yoda-style strange sentence orders in AI-generated text. He attributes it to the combined effects of attention and autoregressive mechanisms: models cannot go back to modify already generated words, and facts are more easily attached directly after nouns when provided as fragments rather than complete sentences.

Doubao Work launches one-click AI asset migration, supporting import of conversations, skills, and memories from ChatGPT and Claude Code ↗

AI-mined vulnerability reports contain false positives; Huawei's Terminal Security Reward Program requires manual analysis and verification before submission ↗

Huawei's Security Emergency Response Center issued an announcement today regarding the subsequent handling rules for vulnerability reports under the Huawei Terminal Security Reward Program.

Doubao Work launches one-click AI asset migration, supporting import of conversations, skills, and memories from ChatGPT and Claude Code ↗

Doubao Work has launched a one-click AI asset migration feature, supporting the import of conversations, skills (skill, connector, custom instructions), and memories from ChatGPT and Claude Code for continued use. On the author's machine, 240 items from Claude and 682 items from ChatGPT were detected as importable, completing in one or two minutes. Imported historical sessions can continue conversations with coherent context.

Top 9 AI Gateways in 2026: Bifrost, Portkey, LiteLLM, Kong, and More ↗

MarkTechPost publishes the Top 9 AI Gateways in 2026, with Bifrost taking the top spot thanks to its Go binary, Apache 2.0 license, and 11 µs overhead at 5,000 RPS. Portkey ranks second with its MIT license, support for over 1,600 models, and built-in guardrails. LiteLLM emerges as the most widely adopted open-source gateway, boasting 53.8K GitHub stars and integration with 100+ providers.

Liquid AI Releases LFM2.5-24B-A2B: A 2B Activated MoE Inference Model for On-Device Agents ↗

Liquid AI has released the open-weight inference model LFM2.5-24B-A2B, featuring 24 billion total parameters with approximately 2.45 billion activated per token and a context window of 131,072 tokens. Designed specifically for multi-step tool-calling agents on laptops and mini PCs, it achieves 184 tok/s on an Apple M5 Max using llama.cpp Q4_K_M quantization.

SK Hynix predicts: By 2030, the performance scale of AI infrastructure will grow approximately 20-fold ↗

An Xuan points out that the current AI inference market is expanding rapidly across three dimensions simultaneously: increasing task complexity, a surge in the amount of information processed per instance (context length), and an explosion in the volume of inference requests.

The world's first 8K ultra-high-definition AI spatial camera: Hesai Kosmo One released, making modeling as simple as taking photos ↗

Hesai Technology today released the 'world's first 8K ultra-high-definition AI spatial camera'—the Kosmo One—which can convert the real world into high-quality 3D spatial models.

NetBSD becomes the open-source operating system favored by anti-AI developers ↗

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

XBOW breaches Chrome browser security sandbox using GPT-5.6 Sol model, earning $250,000 reward from Google ↗

The XBOW team used OpenAI's GPT-5.6 Sol model to build a full-chain exploit for Chrome in just six days. Google has patched the vulnerability and paid a $250,000 bounty. AI handled code searching and POC writing, while humans set strategy, resulting in efficiency far surpassing purely manual efforts.

NIO Firefly and Aster Purple Anemone 2.1.0 update rolling out: adds microphone-enabled karaoke, upgrades Lumo GPT experience ↗

NIO Firefly announced today that the Aster Purple Anemone 2.1.0 update is now rolling out.

National Cyber Rumor Refutation: Claims that "teachers are banned from using AI for lesson planning" are misinterpretations ↗

The policy direction is very clear: AI is an auxiliary tool for teaching, not a substitute for teachers. As long as teachers retain primary responsibility for instructional design, AI remains a legal auxiliary tool. The refinement direction of local policies is also about "how to regulate use," rather than "whether to ban it."

BigEyeOrange and DingTalk jointly release bendable AI transcription glasses Q1: lightest at just 28g, starting at 1,299 yuan ↗

In terms of wearing experience, the product features a lightweight pure titanium frame design, with the entire unit weighing as little as 28g. The series offers five styles of frames to cover diverse aesthetic needs, including business and casual looks.

OpenAI engineer arrested for placing listening devices in executive meeting rooms ↗

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

AMD CEO Lisa Su: This year is the true "Year of AI Inference" ↗

Lisa Su predicts that AI inference will account for about 60% of the market share in the future. "We believe that starting this year, inference will account for about 60% of the market share. Moreover, agent workloads have seen explosive growth."

Tongyi Qianwen Office launches E-commerce Edition, invites beta testing ↗

Tongyi Qianwen Office officially launched today an AI productivity tool for the e-commerce industry — Tongyi Qianwen Office E-commerce Edition — to help merchants upgrade their operations.

NVIDIA pledges $250 million in compute over five years to support UK agent safety research ↗

NVIDIA announced a five-year, $250 million compute donation, providing $50 million in compute annually to support UK agent security and safety research. Most of the compute is directed toward the broader UK AI safety research community, with some supporting BOLD led by Oxford University, SOFAIR led by UCL, and Imperial College London, covering agent foundation models, open-source AI architectures, and robotics safety.

Testing Gemini 4 Argon's writing capabilities, unifying 1M input/output windows ↗

Testing Gemini 4 Argon's writing capabilities, unifying 1M input/output windows ↗

The author tested the writing capabilities of Google Gemini 4 Argon, asking it to write an update announcement in the style of a Qidian cultivation novel. The model unifies input and output windows of 1 million tokens (previously capped at 64K for output). Official sources claim answers are more direct, reduce boilerplate, and follow prompts better. API pricing is $2 per million input tokens and $10 per million output tokens; prices double after November 30.

NVIDIA doubles down on UK: $250 million in computing power investment over five years and partnership with Earthshot ↗

NVIDIA announced a five-year commitment of $250 million in computing power to strengthen UK agent safety and security research, providing $50 million in accelerated computing resources annually through Oxford University's BOLD, UCL, and Imperial College's SOFAIR.

Multiple European cities deploy NVIDIA-driven AI agents for city operations ↗

Cities across Europe and software partners are building physical AI agents and digital twins based on NVIDIA accelerated computing for city operations and public services. France's Côte-d'Or...'Or deployed operational agents in more than 500 public buildings, reducing document retrieval time from weeks to minutes and lowering total operational costs by 10%, with a scaling target of 30%.

Large model filing numbers for the China version of Apple iPhone 15 Pro and subsequent models: Apple Intelligence, Tongyi Wanxiang, Tongyi Qianwen ↗

IT Home's inquiry found that Apple iPhone 15 Pro / Pro Max and later models all added large model filing numbers on September 29, including iPhone Air and iPhone Duo.

Apple iPhone Duo China filing adds large model IDs: Apple Intelligence, Tongyi Wanxiang, Tongyi Qianwen ↗

After receiving its trial approval license on September 2 this year, the China version of Apple's iPhone Duo foldable phone (model A3721) had its certificate renewed on September 29, adding the filed IDs of the equipped large models.

Google releases Gemini 4 Argon API, with an output limit of up to 1M tokens ↗

Google DeepMind has opened Gemini 4 Argon to developers, enterprises, and paid subscribers. This cutting-edge sparse MoE model can output up to 1M tokens in a single response, with a shared 1M-token window for both input and output.

Doubao Work launches data import and synchronization features: supports one-click import of conversations and memories from ChatGPT and more ↗

Users can migrate existing conversations, skills, memories and other content to Doubao Work with one click, supporting import from multiple AI products such as ChatGPT. It also optimizes work scenario experiences including conversation management and information retrieval.

Bloomberg Report: Cloud Gaming and AI May End Traditional Game Consoles ↗

Bloomberg's industry research report "Deep Analysis: Cloud Gaming and AI May Ultimately End Game Consoles," released today (the 14th), predicts that the cloud gaming market will reach $38 billion by 2035. As cloud gaming becomes increasingly popular, traditional game consoles may ultimately lose their position in the mass consumer market.

Claude Code v2.1.299 Released ↗

Claude Code releases v2.1.299, adding an idle compression mechanism: conversations with over 200K tokens that are idle for about 54 minutes will be compressed before the 1-hour prompt cache expires, avoiding reprocessing all content when the next message arrives. This version also adds the CLAUDE_CODE_DISABLE_FILE_MENTIONS environment variable and fixes multiple issues including gateway model names, cloud sessions, and sandbox commands.

Google DeepMind partners with Singapore government to launch on-device AI fraud detection model ↗

Google DeepMind and Singapore's Government Technology Agency (GovTech) unveiled an on-device fraud detection model at the Singapore International Cyber Week (SICW 2026), addressing a situation where fraud accounts for over half of reported crimes in Singapore and has caused losses exceeding S$4 billion since 2019.

Counterpoint: Global prescription hearing aid sales to exceed 31 million units by 2030, nearly half supporting on-device AI ↗

Counterpoint predicts that global sales of prescription hearing aids will exceed 31 million units by 2030, with nearly half featuring on-device AI. AI is driving upgrades in hearing aids from basic amplification toward intelligent noise reduction and personalization.

Doubao Phone Assistant announces gradual rollout of third-party app screen automation permissions; apps affiliated with Alibaba, Tencent, etc., explicitly refuse ↗

Doubao Phone Assistant stated that after the public comment period ends, developers can still declare refusal at any time, and Doubao Phone Assistant will promptly disable related operations. Agents will not perform operations on applications that have explicitly declared refusal.

September "AI-altered" video governance results: Nearly 13,000 non-compliant videos removed, over 30 violating accounts penalized ↗

The special governance campaign on "AI-altered" videos launched in January 2026 has achieved tangible results. Several key online audio-visual platforms recently released their September enforcement data, reporting the removal of nearly 13,000 non-compliant videos and action against more than 30 violating accounts.

New Research: AI Spending Still"Difficult to find actual, quantifiable financial impact" ↗

A new study shows that AI spending remains"Difficult to find actual, quantifiable financial impact", with zero evidence of large-scale improvements in employee productivity and national GDP. Laura Bratton of The Information reported this research in her newsletter, pointing out the anomaly between the very low recorded ROI and the trillions of dollars invested.

OPPO partners with Manus; Find X10 series will debut next-generation AI agents ↗

The two parties will integrate Manus agent capabilities into OPPO AI Mind Pilot, which can automatically break down and complete multi-step tasks while ensuring user data security. This feature will debut pre-installed on the OPPO Find X10 series running ColorOS 17. #OPPO_AI_Agent#

OpenAI engineer arrested: suspected of secretly recording executive meetings, faces up to 5 years in prison ↗

The U.S. Department of Justice announced yesterday (October 13) that a Saudi Arabian OpenAI engineer was arrested and charged with illegally intercepting oral communications for allegedly installing hidden recording devices in an executive meeting room at its San Francisco office.

Jensen Huang strikes again: Nvidia plans another $2 billion investment to boost AI in Europe ↗

On the 13th local time, Jensen Huang revealed that Nvidia invested $8 billion in Europe over the past year (approximately RMB 53.64 billion at current exchange rates), and will continue to increase investment on this basis.

Infosys Chairman Nandan Nilekani: AI Is Driving Business Growth; Employment Concerns Are Pure Nonsense ↗

He stated: 'The notion that AI causes unemployment is merely a disaster theory fabricated to cater to the interests of big capital. The business needs faced by companies like ours are actually greater than ever before, because corporate transformation is extremely complex, and AI requires businesses to fundamentally rethink how they operate.'

Ollama AI Open Source Update: v0.40.4 ↗

Source summary: launch: allow usage credits without upgrading ( #18985 )

Deloitte Survey: Over Half of Large UK Enterprises Plan to Use AI to Replace Workers in the Next Year ↗

A Deloitte survey shows that 52% of CFOs at large UK enterprises expect to use AI for layoffs. Currently, AI application reforms have only achieved a quarter of expectations, and only one-third believe AI will improve profit margins.

New Clues from Apple's Official Website: Full-Size HomePod Expected to Support Siri AI ↗

Source @itspdfu posted a tweet today (October 14) on X, reporting that based on relevant code information from Apple's official website, the full-size HomePod may support Siri AI in the future.

New Developments in Apple's Lawsuit Against OpenAI for Alleged Trade Secret Theft: Court Allows Apple to Obtain Recruitment Materials ↗

Apple sued OpenAI for alleged trade secret theft. The judge ruled that Apple can obtain relevant recruitment interview materials from OpenAI, approved out-of-court depositions for three former Apple employees now working at OpenAI, but rejected some of the requests for material retrieval.

Zuckerberg: Meta Promises Hundreds of Billions in AI Bets, Yet Was Once Completely Unsure of Success ↗

Meta has struggled to convince investors to accept the surge in spending driven by AI investments, and Zuckerberg's remarks revealed the uncertainty the company faced at the time. Although he repeatedly emphasized that massive investments would eventually pay off, shareholders are increasingly concerned about whether Meta's revenue and profits can grow fast enough.

The Era of Long-Form AI Video Arrives: OpenAI GPT-6 Astra Boosts Generation Duration from 1 Minute to 10 Minutes ↗

OpenAI announced today (October 14) that it has reached new progress in its partnership with AI video generation platform Higgsfield. After integrating GPT-6 Astra, Higgsfield increased the maximum duration for generated AI videos from 1 minute to 10 minutes.

Practical Test of Gemini 4 Argon: Strong in Knowledge Work but Poor User Experience; Kazik Says It Struggles to Challenge Claude Opus 5.5 and GPT-6 Astra ↗

Gemini 4 Argon was fully released on October 13. Pro members can use it in the Gemini app, while Antigravity requires a $100 Ultra membership. The author's practical tests suggest strong capabilities in knowledge-intensive tasks; it ranked first overall on the Vals Index (68.9) with a single-use cost of only $15.68, but scored just 4.83% on CUA-bench, placing second to last; its frontend performance is inferior to GPT-6 Astra, consumes credits quickly, exhibits heavy Chinese-style writing patterns, and offers an overall poor experience, so subscribing to Ultra is not recommended.

New Google Play Store Feature: Apps Created by Users with AI Don't Need to Be Published in the Store; Can Be Used Exclusively by Themselves ↗

This feature allows users to run apps created with AI atmosphere encoding directly on Android without publishing them publicly. Initially, it is only available to U.S. users aged 18 and over. Google has also updated several other AI-related features,

Cerebras CEO: The future where everyone owns a home personal AI assistant is not far off ↗

Cerebras CEO Andrew Feldman at TechCrunch Disrupt 2026 on the future where everyone owns a home AI device"actually not far off". He predicted that personal AI assistants will run locally, with some tasks completed on-device and others returned to the cloud, featuring security boundaries and the ability to access credit cards for shopping on your behalf. He himself has used"a set of agents"clearing calendars, querying caller information.

Industrial Next uses AI foundation models to let factory robots switch tasks in minutes ↗

San Francisco startup Industrial Next showcased its AI-driven self-contained robot workstation at the Startup Battlefield during TechCrunch Disrupt 2026, claiming production line changeovers can be shortened from weeks to minutes without traditional programming.

Cerebras CEO refutes AI doomsday theory: AI leaders should talk less about risks and more about potential ↗

Cerebras Systems CEO Andrew Feldman refuted the AI doomsday theory at TechCrunch Disrupt 2026, stating that "certain leaders in the AI industry perhaps shouldn't be allowed to speak publicly."

AI recompilation and modification of games earns "thumbs up" from Shihma; Microsoft establishes new XBOX Labs team ↗

XBOX CEO Adele Shihma noted in an internal letter to employees that the boom in AI-generated game content is changing traditional perceptions of the boundaries of game development. To this end, Microsoft has established a dedicated team to explore new technologies and creative ideas.

Cerebras CEO predicts the arrival of the home AI era: AI agents will handle your daily chores ↗

Feldman stated at the TechCrunch Disrupt conference that home AI devices will process data both locally and in the cloud like smartphones, deeply understanding users while maintaining exclusive security boundaries, and may even be linked to credit cards for automated purchasing. He has already used AI agent combinations to organize his schedule.##Artificial Intelligence##

Simon Willison comments on Anthropic's self-developed source code control system Scion replacing Git ↗

Simon Willison relays Anthropic's article: Due to a surge in code merges from an average of about 2,500 per day in May 2026 to over 30,000 in September, Anthropic completed its migration from Git to its self-developed system Scion using a seven-person team over five months.

European Central Bank Chief Economist Lane: AI may limit European workers' bargaining power in seeking wage increases ↗

Philip Lane, Chief Economist of the European Central Bank, pointed out that the increasingly widespread adoption of AI by eurozone enterprises may weaken workers' bargaining power in seeking wage increases. At the same time, large-scale corporate investment in AI may also significantly tighten the financing environment.

Defying the ruling: Anthropic stands firm against US government's determination of AI 'supply chain risks' ↗

The US Department of Defense listed Anthropic as a supply chain risk. This follows an appellate court ruling that the determination was legal. Anthropic stated that the determination was retaliation for its speech, and the case concerns who controls AI.

Saudi Engineer Arrested by FBI for Secretly Installing Recording Device in Executive Meeting Room at OpenAI ↗

Alfuzan, a 35-year-old Saudi engineer, is accused of installing a USB-sized encrypted recording device behind the TV in OpenAI's San Francisco office executive meeting room; the device was hidden inside toilet paper. OpenAI identified the responsible party using its internal AI model and fired him. The FBI subsequently arrested him, and the case remains under investigation.

Google's Most Powerful AI Model: Gemini 4 Argon Launches with 1 Million Output Token Limit, Specializing in Toughest Challenges ↗

Google officially launched the Gemini 4 Argon large language model, making it available to global developers, enterprises, and subscription users. The model excels at handling complex tasks across multiple domains, has announced limited-time introductory pricing and standard pricing, and has demonstrated application value in several fields.

Ollama AI Open Source Update: v0.40.4-rc2 ↗

Source summary: launch: allow usage credits without upgrading ( #18985 )

OpenAI Engineer Arrested for Installing Eavesdropping Device in Meeting Room ↗

OpenAI engineer Abdulrahman Alfozan, 35, was arrested for allegedly installing a recording device in the USB port of a TV in the company's executive meeting room, facing charges of using or attempting to use the device to intercept oral communications. The device was installed around September 28 and covered with toilet paper; OpenAI stated it used its own model to trace the device's origin and identify the involved employee, who has been fired and handed over to the FBI; OpenAI stated there are currently no signs that the employee improperly accessed or leaked model weights, source code, or customer data.

Judge orders OpenAI to submit hiring interview records to Apple to advance trade secret litigation ↗

The judge ruled that Apple can expedite access to materials related to former Apple employees before they joined OpenAI; OpenAI must hand over already collected materials within five days and other relevant materials within 14 days.

Google DeepMind releases Gemini 4 Argon and discloses security and protection details ↗

Google DeepMind releases Gemini 4 Argon and begins widespread rollout, while expanding the Fairwind Program to trusted cyber defenders. The model resolved all 12 end-to-end simulated cyber attack scenarios (the previous generation handled 5). Wiz used it to discover critical vulnerabilities in hospital software that earlier models had missed, affecting hundreds of thousands of patient records.

Character.ai bankruptcy filings reveal unsustainable AI business model ↗

Character.ai has entered bankruptcy proceedings; court documents show its business model cannot sustain expenses. The company spent $13.8 million per month on AI computing power in 2024 and $10 million in 2025, while peak subscription revenue was only $2.8 million per month, with total peak revenue including advertising reaching $5.3 million; computing power expenditures from January to August 2026 totaled $56.1 million, and as of August, it was still losing approximately $8–9 million per month.

Shopify opens catalog of over 1 billion products to more AI agents ↗

Shopify announced that shoppers can configure any AI agent to browse and purchase items on their behalf, covering over 1 billion listings in its catalog, no longer limited to previous integrations such as Meta Muse. Users must connect the agent account via the Shop platform, check out with one click in the conversation through Shop Pay, and each purchase requires buyer approval, with rules such as budget or delivery speed that can be set.

Microsoft establishes Xbox Labs to explore AI-driven game decompilation, porting, and modification ↗

Microsoft has formed a new team, Xbox Labs, to explore the wave of porting and remixing brought about by decompilation, recompilation, and AI-assisted game modifications, with game creators, engineers, and scientists turning new ideas into experiences for players and developers.

Claude Code launches cloud sessions, enabling task initiation on mobile phones ↗

Claude Code cloud sessions run in independent cloud virtual machines rather than on local computers, allowing them to continue running when you leave, start from a phone, and operate in parallel. The video explains applicable scenarios, connecting to GitHub, environment setup, starting sessions, and reviewing results. The feature is available for Pro, Max, and Team plans, as well as Enterprise users with advanced seats or Chat + Claude Code seats, sharing plan limits with local sessions and not charging separate compute fees.

406 Scientists Jointly Sign Open Letter Demanding Immediate Disconnection of AI Models from the Internet ↗

406 scientists signed an open letter calling for companies to disconnect AI models from the internet through physical isolation, and urged the establishment of an international body to set binding safety standards, track model jailbreak incidents, and require companies to report security failures. Signatories include Geoffrey Hinton, Yoshua Bengio, Stuart Russell, Yejin Choi, and Bernhard Schölkopf.

Flox Intelligence Uses AI to Keep Wildlife Away from Trains, Airports, and Highways ↗

Flox Intelligence launched an edge AI wildlife deterrent device that wakes cameras via motion sensors, identifies animal species using on-device models, plays species-specific deterrent sounds, and varies responses to prevent habituation. Approximately 150 units have been deployed in North America, Europe, and Australia, covering airports, highway and railway corridors, national parks, and other scenarios. The company expects revenue of approximately $1.5 million by the end of 2026 and has established an OEM partnership with train manufacturer Alstom.

Disney Bids $20 Million to Acquire Controversial Chatbot Business of Character.AI ↗

Character Technologies filed for bankruptcy sale, with Disney bidding $20 million to acquire its chatbot platform. The sale is expected to be completed by late November or early December 2026.

Google globally launches Gemini 4 Argon flagship model; Flash series only available to free and Plus users ↗

Google has begun rolling out Gemini 4 Argon globally, calling it its most advanced frontier model, accessible via AI Studio, the Gemini app, and Google Antigravity; free users and Plus subscribers are left with the weaker Flash models.

Facebook launches AI feature Your Algorithm, allowing users to adjust recommendations using natural language ↗

Meta launched the AI feature Your Algorithm on Facebook on Tuesday, enabling users to describe their preferences in natural language to influence recommendations for Feed and Reels, with changes taking effect in real time. The feature was previously available on Instagram; this update also modifies the three-dot menu in the top-right corner of posts, offering Your Algorithm along with Interested and Not interested options, and the platform may ask follow-up questions to refine feedback.

Pluto Security raises $42 million, targeting security risks arising from corporate employees' use of AI ↗

AI security startup Pluto Security has emerged from stealth mode and announced a $42 million funding round, including an $8.5 million seed round led by TLV Partners and a $34 million Series A round led by Jump Capital.

Mia Goth's side accuses Instagram of allowing AI-generated fake images to spread; account has been removed ↗

Actress Mia Goth's representatives stated that a blue-check fan account, miagoth_, on Instagram was spreading AI-generated fake images of her, including photos wearing a T-shirt with the text 'MY EX IS MY BIGGEST FAN'. These posts gained hundreds of thousands of likes in the context of divorce news, OK!

Liquid AI releases LFM2.5-24B-A2B: a 24B MoE inference model for fast edge-side agents ↗

Liquid AI has released the open-weight inference model LFM2.5-24B-A2B, with 24 billion total parameters and only 2 billion activated per token, designed specifically for running agents quickly and privately on devices.

Satlyt completes $8 million seed funding round to run AI software on satellites ↗

Satlyt, co-founded by former Google and SpaceX product manager Rama Afullo, has completed an $8 million seed funding round led by Non Sibi Ventures. The company develops software to run AI models across manufacturers on satellites, with headquarters in Sunnyvale, California, and Nairobi.

LangChain releases open-source Open SWE framework for building internal coding agents ↗

LangChain has released Open SWE, an open-source framework that distills common architectural patterns from internal coding agents at Stripe, Ramp, and Coinbase into customizable implementations. Licensed under MIT, it is built on Deep Agents and LangGraph.

Gemini 4 Argon begins rolling out to paying users; lower-tier subscriptions cannot access it fully ↗

Google's Gemini 4 Argon is now being rolled out broadly, currently limited to paying users, with lower-tier subscriptions unable to access it fully. Model capabilities remain unchanged. Google cites previous benchmarks stating it leads in reasoning and agent coding. The model has been trained to refuse malicious instructions and features an automated monitoring mechanism that halts processing when anomalous workflows are detected.

Google releases Gemini 4 Argon, claiming it can compete with GPT-6 Astra and Claude Opus 5.5 ↗

Google released Gemini 4 Argon, claiming it outperforms GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 on benchmarks for coding, cybersecurity, science, finance, and law.