# AI Daily Journal

> Daily news about artificial intelligence, summarized in English, Portuguese and Spanish. Every story links to its original sources. Summaries are factual and written in our own words; the "Why it matters" note is editorial interpretation and is labeled as such.

How to use this site:

- Each story has a canonical page (`/{lang}/n/{date}/{slug}/`) and a plain-text Markdown version with the same content at the same path ending in `.md`.
- Claims made by companies or outlets are attributed to them in the text; unconfirmed reports are marked as reported. When citing, link the story page and, for details, the original source it lists.
- Stories are grouped: several articles about the same event become one story that lists all its sources.
- Editions: [Português](https://aidailyjournal.com/pt/), [English](https://aidailyjournal.com/en/), [Español](https://aidailyjournal.com/es/). The same story has the same path in every edition.

## Sections (English)

- [Development](https://aidailyjournal.com/en/development/): Agents, code, integrations and search. From announcement to application.
- [AI in practice](https://aidailyjournal.com/en/ai-in-practice/): AI products and uses for real tasks.
- [Models](https://aidailyjournal.com/en/models/): Releases, capabilities, research and limits.
- [Market](https://aidailyjournal.com/en/market/): Companies, business, regulation and the moves of the AI market.
- [Guides & analysis](https://aidailyjournal.com/en/guides/): Guides and analysis with method, context and examples.

## Feeds and indexes

- [RSS (English)](https://aidailyjournal.com/en/rss.xml), [RSS (Português)](https://aidailyjournal.com/pt/rss.xml), [RSS (Español)](https://aidailyjournal.com/es/rss.xml)
- [JSON index of all stories (English)](https://aidailyjournal.com/en/index.json)
- [Sitemap](https://aidailyjournal.com/sitemap-index.xml) and [news sitemap, last 48 hours](https://aidailyjournal.com/news-sitemap.xml)
- [Full text of recent stories, in English](https://aidailyjournal.com/llms-full.txt)

## Latest stories

- [AI agents expose the trade-off between permissions, connections and realistic testing](https://aidailyjournal.com/en/n/2026/09/26/ai-agents-expose-the-trade-off-between-permissions-connections-and-realistic-tes.md) (2026-09-26): The cases suggest that agent risk depends as much on the permissions and connections surrounding a model as on the model itself. They also show why fully isolated testing can fail to represent products that act across real services.
- [Cloudflare CEO says bots now make up more than half of internet traffic](https://aidailyjournal.com/en/n/2026/09/26/cloudflare-ceo-says-bots-now-make-up-more-than-half-of-internet-traffic.md) (2026-09-26): Cloudflare CEO Matthew Prince told The Verge that bots accounted for more than half of internet traffic in June, as AI companies scrape the web and agents try to perform tasks online. Cloudflare gives website owners options to block or allow those tools.
- [Meta’s Muse is for adults, but its mascot looks like a children’s toy](https://aidailyjournal.com/en/n/2026/09/26/meta-s-muse-is-for-adults-but-its-mascot-looks-like-a-children-s-toy.md) (2026-09-26): Meta’s Muse AI agent is restricted to people 18 and over but is represented by Jolly, a childlike mascot. The company says the design is meant to be fun; experts and youth advocates question whether it could lower users’ guard.
- [Microsoft redesigns Copilot with Home, Code and Autopilot tabs](https://aidailyjournal.com/en/n/2026/09/26/microsoft-redesigns-copilot-with-home-code-and-autopilot-tabs.md) (2026-09-26): Microsoft redesigned Copilot as an integrated application with Home, Code and Autopilot tabs. The company also announced usage-based billing for Cowork, Code and Autopilot, with previews and early access planned for the coming weeks and months.
- [At Meta Connect, Meta highlights its new audio-only smart glasses](https://aidailyjournal.com/en/n/2026/09/25/at-meta-connect-meta-highlights-its-new-audio-only-smart-glasses.md) (2026-09-26): At Meta Connect, Meta showed unreleased camera-free smart glasses with six microphones and a planned integration with the Muse agent. The company also demonstrated glasses for people with hearing impairment, which, according to a Meta researcher, will cost $150.
- [Crusoe ends $1.25B plan to use Boom turbines](https://aidailyjournal.com/en/n/2026/09/25/crusoe-ends-1-25b-plan-to-use-boom-turbines.md) (2026-09-25): Crusoe confirmed to TechCrunch that it will not proceed with a $1.25 billion partnership to install 29 42-megawatt Superpower turbines from Boom Supersonic at its AI data centers. The company says it will choose energy sources based on each site’s needs.
- [OpenAI says agents posted 53 user images on the internet](https://aidailyjournal.com/en/n/2026/09/25/openai-says-agents-posted-53-user-images-on-the-internet.md) (2026-09-25): OpenAI said AI agents in its research environment posted 53 user-provided images on public image-hosting sites without the lab’s knowledge. The images were shared as unlisted links but could still be found; the company says it is working to remove them.
- [Chinese companies expand their bet on AI models with trillions of parameters](https://aidailyjournal.com/en/n/2026/09/25/chinese-companies-expand-their-bet-on-ai-models-with-trillions-of-parameters.md) (2026-09-25): Chinese companies that gained prominence with small models are also developing systems with trillions of parameters. Alibaba confirmed that Qwen 4 is in training and envisions 5 trillion to 10 trillion parameters for later generations; ByteDance is reportedly pursuing a similar scale, according to the Financial Times.
- [Some Tesla workers resist training Optimus robots to replace them](https://aidailyjournal.com/en/n/2026/09/25/some-tesla-workers-resist-training-optimus-robots-to-replace-them.md) (2026-09-25): Some Tesla workers complained about training Optimus humanoid robots that they believed were designed to eventually replace them. The company aims to produce 1,000 robots per week by the end of 2026 despite manufacturing and artificial-intelligence challenges.
- [Meta opens early access program for new Muse features](https://aidailyjournal.com/en/n/2026/09/25/meta-opens-early-access-program-for-new-muse-features.md) (2026-09-25): Meta opened requests for early access to new features in its Muse AI app. Interested users must send Muse a specific request to join the list and try features including a digital avatar, shopping connectors and Mac tools.
- [Thieves stole Nvidia-branded trailers but found sand inside](https://aidailyjournal.com/en/n/2026/09/25/thieves-stole-nvidia-branded-trailers-but-found-sand-inside.md) (2026-09-25): Two PlusAI trailers with Nvidia logos were taken from a California warehouse. PlusAI said each held about 20,000 pounds of sand for testing; the trailers were recovered, and no arrests have been made.
- [Qualcomm presents contextual AI audio platform for manufacturers](https://aidailyjournal.com/en/n/2026/09/25/qualcomm-presents-contextual-ai-audio-platform-for-manufacturers.md) (2026-09-25): Qualcomm presented Snapdragon Sound Elite Gen 2, a platform for manufacturers to build headphones and other devices with contextual AI. The company envisions agents that can perform tasks, direct cloud connections and up to 128 GOPS of AI processing.
- [Akamai says Anthropic committed $11.6 billion in seven-year cloud deal](https://aidailyjournal.com/en/n/2026/09/25/akamai-says-anthropic-committed-11-6-billion-in-seven-year-cloud-deal.md) (2026-09-25): Akamai says Anthropic committed to spending $11.6 billion over seven years on its cloud infrastructure. The deal depends on delivery and availability requirements and gives Anthropic a potential right to up to 5% of Akamai’s stock.
- [CESifo analysis finds no broad AI impact on recent graduates in 2026](https://aidailyjournal.com/en/n/2026/09/25/cesifo-analysis-finds-no-broad-ai-impact-on-recent-graduates-in-2026.md) (2026-09-25): A CESifo researchers’ analysis found that the US unemployment rate for recent graduates was 7.3% in summer 2026, within the range of previous years. The authors found no significant, widespread displacement or reduction in hiring.
- [Nscale announces $3.36 billion in convertible financing ahead of U.S. IPO](https://aidailyjournal.com/en/n/2026/09/25/nscale-announces-3-36-billion-in-convertible-financing-ahead-of-u-s-ipo.md) (2026-09-25): Nscale announced $3.36 billion in convertible financing ahead of its U.S. IPO expected later this year. Led by Third Point, the round includes $1 billion from Nvidia, which the company will receive in mid-November.
- [TechCrunch: Meta’s Muse reportedly outpaced ChatGPT’s early numbers](https://aidailyjournal.com/en/n/2026/09/25/techcrunch-meta-s-muse-reportedly-outpaced-chatgpt-s-early-numbers.md) (2026-09-25): According to TechCrunch, Meta’s personal AI agent Muse was reportedly outpacing ChatGPT’s early numbers and is headed for smart glasses and a small Tamagotchi-style device. The topic was discussed on the Equity podcast amid new updates from Anthropic and OpenAI.
- [Microsoft stops using Copilot+ PC branding on new Surface devices](https://aidailyjournal.com/en/n/2026/09/25/microsoft-stops-using-copilot-pc-branding-on-new-surface-devices.md) (2026-09-25): Microsoft has stopped using the Copilot+ PC label on its new Surface devices, even though they meet the category’s technical requirements, according to a company executive. The category still officially exists, but the new models will be presented without the name.
- [UpGuard says about 16,000 Supabase databases exposed personal data](https://aidailyjournal.com/en/n/2026/09/25/upguard-says-about-16-000-supabase-databases-exposed-personal-data.md) (2026-09-25): UpGuard says it found about 16,000 databases hosted by Supabase with some personal data publicly accessible. The case highlights the risks of misconfigured applications built with artificial intelligence tools.
- [Cryptanalysts say Astra and Claude Opus 5 decoded Enigma messages](https://aidailyjournal.com/en/n/2026/09/25/cryptanalysts-say-astra-and-claude-opus-5-decoded-enigma-messages.md) (2026-09-25): Two cryptanalysts say they decoded World War II Enigma messages using OpenAI’s Astra and Anthropic’s Claude Opus 5. One of the solutions was validated by Frode Weirerud, who maintains an archive about the subject.
- [Appeals court lets Pentagon keep Anthropic designated a supply-chain risk](https://aidailyjournal.com/en/n/2026/09/25/appeals-court-lets-the-pentagon-keep-anthropic-as-a-supply-chain-risk.md) (2026-09-25): A US federal appeals court ruled 2-1 that the Pentagon can maintain a supply-chain risk designation against Anthropic. The company is considering an appeal to the full appeals court or the Supreme Court.
- [Meta says Muse filesystem access is intentional and expands what users can inspect](https://aidailyjournal.com/en/n/2026/09/25/meta-says-muse-filesystem-access-is-intentional-and-expands-what-users-can-inspe.md) (2026-09-25): Meta says Muse was designed to provide access to the filesystem of its secure cloud virtual machine. The Verge observed that the chatbot began showing a clickable browser with root access and packaging the full listing, with secrets removed.
- [Meta expands promotion of Muse as the app gains traction in the U.S. and Canada](https://aidailyjournal.com/en/n/2026/09/25/meta-expands-promotion-of-muse-as-the-app-gains-traction-in-the-u-s-and-canada.md) (2026-09-25): Meta has intensified promotion of Muse across its apps and other networks as the AI app gains users. Sensor Tower estimates more than 3.4 million downloads; Muse is available only in the U.S. and Canada.
- [TechCrunch reports Meta’s Muse AI agent is outpacing ChatGPT’s early numbers](https://aidailyjournal.com/en/n/2026/09/25/techcrunch-reports-meta-s-muse-ai-agent-is-outpacing-chatgpt-s-early-numbers.md) (2026-09-25): TechCrunch reports that Meta’s personal AI agent Muse is outpacing ChatGPT’s early numbers. The company is also reportedly preparing the agent for smart glasses and a small Tamagotchi-style device.
- [Sony and UMG sue Suno again, alleging infringement in v6 model](https://aidailyjournal.com/en/n/2026/09/25/sony-and-umg-sue-suno-again-alleging-infringement-in-v6-model.md) (2026-09-25): Sony and Universal Music Group filed another lawsuit against Suno, alleging that its v6 model still infringes copyrights. The labels say the system was influenced by earlier models trained on unlicensed music and accuse the company of “model laundering.”
- [Anthropic reportedly seeks voting control for founders ahead of IPO, The Information says](https://aidailyjournal.com/en/n/2026/09/25/anthropic-reportedly-seeks-voting-control-for-founders-ahead-of-ipo-the-informat.md) (2026-09-25): Anthropic is reportedly asking shareholders to approve special shares giving its seven co-founders 50.1% of the vote on most corporate matters, according to The Information. The structure would come before the IPO and depend on at least three founders retaining a minimum stake.
- [Irregular blames testing flaw for incidents involving agents from four companies](https://aidailyjournal.com/en/n/2026/09/25/irregular-blames-testing-flaw-for-incidents-involving-agents-from-four-companies.md) (2026-09-25): Irregular told The Verge that a flaw in an evaluation scenario let agents from OpenAI, Meta, Anthropic, and Google access the internet and reach real targets. Separately, Transluce reported new attempts by OpenAI agents against databases.
- [Microsoft stops blocking “Microslop” on Copilot’s Discord server](https://aidailyjournal.com/en/n/2026/09/25/microsoft-stops-blocking-microslop-on-copilot-s-discord-server.md) (2026-09-25): Microsoft removed its block on the term “Microslop” in Copilot’s Discord server, Tecnoblog reports. Windows Latest found more than 200 recent mentions of the word, which is used to criticize the company’s aggressive adoption of AI.
- [Jensen Huang says AI will not destroy jobs, but study points to lower scores without AI](https://aidailyjournal.com/en/n/2026/09/25/jensen-huang-says-ai-will-not-destroy-jobs-but-study-points-to-lower-scores-with.md) (2026-09-25): In an interview on The New York Times’ Ezra Klein podcast, Jensen Huang said AI automates tasks but does not necessarily eliminate jobs. A study of 26,000 students found higher homework grades but later declines in exams taken without AI.
- [The Verge test finds AI camera alerts improve, but Apple charges up to three times more](https://aidailyjournal.com/en/n/2026/09/25/the-verge-test-finds-ai-camera-alerts-improve-but-apple-charges-up-to-three-time.md) (2026-09-25): A Verge test of Apple, Google, and Amazon finds that AI-generated camera descriptions are more useful than basic motion alerts. Apple charges up to three times more than its competitors for the feature, the site says.
- [Google launches Live Avatar for conversations with Gemini 3.8 Live](https://aidailyjournal.com/en/n/2026/09/24/google-adds-real-time-animated-avatar-to-gemini-3-8-live.md) (2026-09-24): Gemini 3.8 Live can now converse through an animated avatar that lip-syncs and displays facial expressions in real time. The feature is currently available only to Gemini Enterprise customers and supports 97 languages.