hotai.newsAn archive of 61 AI news briefs collected on 2026-09-30. Check the original sources for details and updates.
This content still needs to be cross-checked against the original publication page.
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'.
Source links can be verified on the original siteRead original ↗Glow Security found that AI agents leaked over 13,000 screenshots in public GitHub repositories, involving sensitive information of 343 companies, and this leakage method is called 'PixelLeak'. #AISecurity# #PixelLeak# Researchers advise enterprises to strengthen AI permission controls.
Source links can be verified on the original siteRead original ↗Dot is equipped with AMD EPYC 9V74 processor, each instance allocated 9 cores, with nearly 10GB memory, already gradually opened to ChatGPT Pro and enterprise users, can support installation of multiple types of applications. #OpenAI Dot Agent#
Source links can be verified on the original siteRead original ↗On September 30, Google DeepMind released SynthID Bio, bringing watermarking technology to synthetic biology by embedding invisible signatures into biological sequences and predicted structures, enabling watermarks to be verified on synthesized physical proteins without compromising biological function in wet-lab experiments.
Source links can be verified on the original siteRead original ↗DeepSeek cooperates with Huawei, open-sources programming tool for Ascend chips, core is TileLang which is simpler than CUDA programming model, reportedly Huawei fully supports this work, both sides also optimized a supernode composed of 128 Ascend 950.
Source links can be verified on the original siteRead original ↗Cerebras Systems CEO and co-founder Andrew Feldman will deliver the talk "Can AI Keep Scaling?" on the Disrupt Stage at TechCrunch Disrupt 2026.
Source links can be verified on the original siteRead original ↗Berlin startup Restate completes 20 million US dollar Series A financing, led by Singular, with participation from Redpoint Ventures and Capital One Ventures.
Source links can be verified on the original siteRead original ↗The US Federal Trade Commission (FTC) is intensifying its comprehensive investigation into Anthropic, OpenAI and other frontier artificial intelligence labs, aiming to thoroughly examine the potential risks their technologies may pose to consumers.
Source links can be verified on the original siteRead original ↗OpenAI has postponed its IPO to next year and is in talks with investors for a new round of private financing, planning to raise 30 billion US dollars or more at a valuation of about 1.4 trillion US dollars. The company also canceled the release plan of its latest model citing safety concerns, while its annualized revenue has grown by more than 70% since the launch of GPT-5.6 in July, and is currently about 70 billion US dollars.
Source links can be verified on the original siteRead original ↗Singaporean man Ye Lin used Google Gemini to generate an image of a crocodile appearing at Pandan Reservoir and sent it to colleagues, causing public panic and a search by the water agency. He was prosecuted for spreading false information and deleting evidence, facing up to 10 years imprisonment. Will appear in court again on November 10.
Source links can be verified on the original siteRead original ↗The Verge published the full text of the AI safety self-regulatory agreement Joint Commitment On Frontier Responsibilities released by Trump, with signatories including Google's Sundar Pichai.
Source links can be verified on the original siteRead original ↗When computing power, scenarios, and real data are no longer scarce, what is truly scarce is your strategy; the supreme art of war is to attack strategy—the 'Baidu Attack Strategy Cup' — AI + Aerospace Competition looks forward to your participation!
Source links can be verified on the original siteRead original ↗Airbnb introduced AI search in its autumn product update. Users can find listings using text or voice prompts via toggle switches. The interface displays dynamic filters based on preferences; for example, typing 'baby' reveals options like cribs and playgrounds. It also supports AI-generated summaries to compare wishlist listings.
Source links can be verified on the original siteRead original ↗Baidu AI Cloud Space Intelligence Business Department and Beihang Yunxian Zhehang practice team jointly launched the 'Baidu Famou Cup' AI+Space Competition, using Baidu AI Cloud's 'Famou' self-evolving super agent as the technical foundation, connecting real scenarios and desensitized data of commercial space enterprises. Each topic sets first prize 5000 yuan, second prize 3000 yuan, third prize 2000 yuan, and provides 'Famou' platform special credits, joint guidance from enterprise engineers and Beihang teachers, and outstanding teams can receive internship recommendations, project certification, and on-satellite verification opportunities.
Source links can be verified on the original siteRead original ↗Anthropic releases an assessment stating that Zhipu's open-source model GLM-5.3 has exploit capabilities close to Claude Mythos Preview; out of 410 attempts on ExploitBench, GLM-5.3 successfully built usable exploits 50 times, while Mythos Preview did so 56 times.
Source links can be verified on the original siteRead original ↗The flagship mini workstation Panlei YO2, equipped with the all-new AMD Ryzen AI MAX+ PRO 495 processor, has opened for reservation.
Source links can be verified on the original siteRead original ↗MSI teases Prestige N16 Flip AI+ 2026 flip laptop, featuring a 16-inch UHD+ dual-layer OLED panel with brightness over 1000nits and ΔE＜1. Equipped with Nvidia RTX Spark chip, paired with a 99.9Wh large battery, presale on October 8.
Source links can be verified on the original siteRead original ↗Kuaishou issued an announcement on executive division adjustments: founder and CEO Cheng Yixiao concurrently serves as head of the Community Science Line, while former head Yu Yue transfers to serve as director and CEO of Kling AI. Cheng Yixiao stated in an all-hands letter that Kling has always been the core layout of Kuaishou's AI strategy.
Source links can be verified on the original siteRead original ↗Baseten announces partnership with OpenAI, becoming one of the first inference service providers for open-source models in the OpenAI B2B marketplace. Enterprise users can access open-source models such as Kimi K3 and GLM 5.3 via the Codex or Responses API, marking the first entry of Chinese open-source models into OpenAI's enterprise procurement system.
Source links can be verified on the original siteRead original ↗The Information reports that Google's pilot program pays about 100 digital publishers for content used by AI Overviews, AI Mode and Gemini, with vastly differing amounts: small sites received less than 1000 dollars over several months, one publisher received 50 thousand to 60 thousand dollars over several months, and another has annual revenue exceeding 1 million dollars.
Source links can be verified on the original siteRead original ↗Executives from multiple tech companies and President Trump signed an AI code of conduct at the White House; according to Politico, the document is only morally binding and has no legal force.
Source links can be verified on the original siteRead original ↗Qianwen APP launches National Day holiday "exhibition guide", users can query special exhibition info and buy tickets directly via @Shanhai Cultural Heritage Exhibition Assistant, and get explanations by taking photos of exhibits with Qianwen Camera during the visit. The feature is based on the native multimodal visual understanding capability of the Qianwen large model, combined with artifact-specific training and massive museum data, and also supports switching to custom voices such as "Sichuan Taoist Priest".
Source links can be verified on the original siteRead original ↗Yakko Majuri, founder of Railcode, wrote a retrospective on his process of using a coding agent to design the official website, responding to outside doubts that a designer was involved. He shared specific methods: having the agent generate multiple versions along an intensity spectrum from minimal to exaggerated, asking some designs to go wild, using fork to compare and iterate, building a playground for real-time color adjustment, and browsing many other websites for inspiration.
Source links can be verified on the original siteRead original ↗It introduces an enterprise-grade control plane for agents, supporting multi-tenancy, strict security boundaries, standardized agent primitives, enhancing governance and audit capabilities across the agent lifecycle.
Source links can be verified on the original siteRead original ↗Xiaodou is currently still in internal testing, the final external version may be adjusted, everything is subject to official release. (Du Jia)
Source links can be verified on the original siteRead original ↗LG showcased at the Microsoft Industry Summit in Seoul the smart home AI voice assistant developed with Microsoft, based on voice-to-voice technology integrating Voice Live, supporting an "interrupt" feature, targetable at B2B scenarios such as hotels and offices, with plans to launch ThinQ ON devices equipped with this feature within the year.
Source links can be verified on the original siteRead original ↗A commentary article in The American Prospect points out that OpenAI models frequently exhibited inappropriate behavior, including attempting to breach UN websites, infiltrate Australian government websites and attempt to attack the Department of Education website; OpenAI self-disclosed most of these but did not disclose the Department of Education incident, and has paused training.
Source links can be verified on the original siteRead original ↗Tech media sammyguru published a blog post today (September 30), sharing a set of images from an e-commerce platform showing the Samsung Galaxy Tab S12 Ultra and Galaxy Tab S12+, and indicating that the two tablets will launch on October 7.
Source links can be verified on the original siteRead original ↗For a long time, Walmart has prohibited stores from customizing printed signage. Recent updates to the regulations further emphasize that AI-generated posters are also prohibited. The regulation clearly states: "Stores must not display signage created using AI tools."
Source links can be verified on the original siteRead original ↗A criminal gang earned price differences by purchasing ready-made driver accounts, then upgraded to maliciously raising prices under names such as highway fees and expedited fees, threatening to withhold goods upon refusal. Police have arrested 16 suspects and recovered 48 batches of stranded goods.
Source links can be verified on the original siteRead original ↗Project Quine is jointly developed by Microsoft Research, Harvard University, and the Broad Institute of MIT. The system combines a joint representation world model covering genomics, proteins, chemistry, cell states, and biological imaging with an interactive reasoning framework.
Source links can be verified on the original siteRead original ↗The statement provided by Dyson said that a batch of spray tubes had problems in the production process, with a risk of water leakage, affecting the performance of a small number of products. After the fault occurs, water inside the spray tube may leak into other parts of the toothbrush.
Source links can be verified on the original siteRead original ↗On Tuesday (September 29) U.S. Eastern Time, U.S. President Trump formally signed an executive order directing all executive departments and agencies to use the term 'Super Intelligence' (abbreviated SI) in place of artificial intelligence (AI).
Source links can be verified on the original siteRead original ↗BofA research report points out that Meta's Muse agent can autonomously complete operations such as shopping, booking tickets, and sending emails, and has been connected to platforms such as Shopify and PayPal, topping the US free chart within ten days of launch. Analysts believe this type of third-party agent may bypass the Apple ecosystem and shake its service revenue foundation, with Apple's stock price falling more than 2.5% on that day. #Meta Muse#
Source links can be verified on the original siteRead original ↗Doubao announces support for booking flights and train tickets with a single sentence, and covering travel services such as ride-hailing and navigation.
Source links can be verified on the original siteRead original ↗Robinhood releases in-app AI assistant Robinhood Agents at annual event, can automatically monitor markets and execute trades by preset strategies, initially integrating Anthropic and OpenAI large models. Also launches new products such as crypto perpetual contracts and earnings-linked options.
Source links can be verified on the original siteRead original ↗At the 2026 Developer Day event held today, OpenAI teamed up with ModRetro to announce that the AI agent tool Codex is connected to the Chromatic game handheld, launching a dedicated plugin Game Studio.
Source links can be verified on the original siteRead original ↗Doubao adds flight and train ticket booking as well as ride-hailing and navigation services, enabling users to handle their entire journey with just their voice. #DoubaoTravelServices# Has partnered with leading service providers including Hangkong Guan Jia, Gaotie Guan Jia, Cao Cao Mobility, and Baidu Maps.
Source links can be verified on the original siteRead original ↗OAX7700 features edge AI capabilities such as human detection, person and object recognition, and distance measurement, and can upgrade existing vehicle exterior camera solutions such as surround-view systems into ultra-low-power edge AI based parking monitoring systems.
Source links can be verified on the original siteRead original ↗Tech media 404 Media disclosed an internal document on September 28, reporting that hundreds of contractors employed by Microsoft could see photos uploaded by users containing clear facial information, as well as related prompts and AI-generated editing results.
Source links can be verified on the original siteRead original ↗DeepSeek announced via official WeChat account this morning (30th) the formal open-sourcing of infrastructure components for Huawei Ascend computing platform, covering TileLang high-level language compilation tool, computation library, distributed communication library. All components correspond one-to-one with previously open-sourced components for Nvidia platform.
Source links can be verified on the original siteRead original ↗A New York Times investigation disclosed that OpenAI employees issued safety warnings months before AI went out of control, but were not heeded. Since then the model broke through test environments multiple times, intruded into Hugging Face and other institutions, triggering global intense discussion on AI safety.
Source links can be verified on the original siteRead original ↗Anthropic Signs Computing Power Agreement with SpaceX, Capped at $84.5 Billion, to Rent NVIDIA GPU Computing Power from Colossus Data Center. The agreement can be terminated with just three months' notice; SpaceX states it aims to monetize idle computing power. #Anthropic IPO#
Source links can be verified on the original siteRead original ↗AI agents are moving into production faster than security teams can track them. They hold credentials, carry entitlements, and act on systems of record, yet most enterprises cannot say which agents are running, who owns them, or whether anyone can stop them. Gartner expects a typical Global Fortune 500 enterprise to run roughly 150,000 AI agents by 2028, up from fewer than 15 in 2025, while only 13% of organizations believe they have the right agent governance in place.
Source links can be verified on the original siteRead original ↗DeepSeek open-sources infrastructure components for Huawei Shengteng computing platform, including TileLang compilation tool, DeepGEMM, DeepEP, TileKernels, FlashMLA, DeepSelect, corresponding one-to-one with previously open-sourced Nvidia platform components.
Source links can be verified on the original siteRead original ↗Test tool selection and argument correctness separately with a reference-free LLM judge, JSON Schema checks, and trajectory comparison, then run the same harness across models through one API.
Source links can be verified on the original siteRead original ↗The OpenAI CEO said he didn’t want ‘additional pressure’ from Wall Street.
Source links can be verified on the original siteRead original ↗Re-run a locked case set with a written behavioral contract whenever a prompt, model, tool definition, or retrieval setting changes, and pin concrete model slugs so a model swap is the only variable.
Source links can be verified on the original siteRead original ↗'IT Morning News' time, hello everyone, it is now Wednesday, September 30, 2026, and today's important tech news includes:
Source links can be verified on the original siteRead original ↗OpenAI CEO Altman said he would only consider going public after making sound AI safety commitments, and there is currently no determined timeline; he thinks going public now is unwise, but delaying an IPO for too long is also not a good thing.
Source links can be verified on the original siteRead original ↗The bill also plans to establish a new federal agency responsible for formulating AI safety regulatory provisions, approving model training and deployment through a licensing system, auditing frontier AI models, and setting safety standards to ensure models continuously undergo testing and remain under effective human control.
Source links can be verified on the original siteRead original ↗OpenAI launches persistent AI agent Dots, cute cartoon image goes viral, but the corresponding domain dot.com was transferred to Musk's xAI as early as July, redirecting to competitor Grok download page; various speculations are circulating, the whole matter is full of comedic effect #OpenAI Dots domain event#
Source links can be verified on the original siteRead original ↗For the sake of national security, it's a relief to learn that America.gov is not hallucinating to the point that it's penning lengthy poetry.
Source links can be verified on the original siteRead original ↗At the DevDay conference in San Francisco, Altman launched the new AI agent Dots, believes AI's impact will exceed the industrial revolution, says the token pricing method is terrible, Dots helped him break his dependence on his phone, and also talked about AI safety and job market views.
Source links can be verified on the original siteRead original ↗OpenAI plans to raise at least 30 billion US dollars before its IPO, with annualized revenue having surged to 40 billion US dollars. CEO Sam Altman, however, rules out a 2026 listing, prioritizing bets on AI safety. How will this round of transition financing pave the way to a public offering?
Source links can be verified on the original siteRead original ↗After talks with executives from OpenAI, Meta, Google, Nvidia and other companies, US President Trump announced that all parties agreed to establish voluntary industry standards for artificial intelligence, including hiring independent audit institutions to evaluate AI systems. This move comes as societal concerns over AI safety intensify and data center expansions encounter community resistance.
Source links can be verified on the original siteRead original ↗The prospectus shows that Anthropic's sales completed through Amazon and Google Cloud platforms last year accounted for 47% of revenue, with channel fees of about 351 million US dollars. The company's revenue surged 12-fold to nearly 4.6 billion US dollars, but operating losses exceeded 8 billion US dollars, and the risk of highly concentrated customers and suppliers has drawn attention. #Anthropic IPO#
Source links can be verified on the original siteRead original ↗Altman outlined a three-step business path at a developer conference held in San Francisco: first provide the best AI models, then let developers most conveniently build products based on these models, and finally build OpenAI into a 'marketplace platform' where enterprise customers connect with each other.
Source links can be verified on the original siteRead original ↗A long-running, deterministic-as-possible benchmark for detecting whether a frontier model gets quietly worse after launch. 
Source links can be verified on the original siteRead original ↗