AI 日报

AI 日报 · 2026-09-26

这期日报从 44 条资讯中筛选出 14 条重点 AI 新闻。关注主题集中在 ai-governance、anthropic、cloudflare。如果只先读两条，可以从《Trump admin using AI to deny medical care for seniors in disastrous experiment》、《Anthropic signs $11.6 billion cloud deal with Akamai, pushing its compute spending past $500 billion in under a year》开始。

Overview

当天导读

从 44 条资讯中筛选出 14 条

Trump admin using AI to deny medical care for seniors in disastrous experiment

This is a high-impact report on the real-world deployment of AI in Medicare prior

关联报道 1

Anthropic signs $11.6 billion cloud deal with Akamai, pushing its compute spending past $500 billion in under a year

A potentially major development in AI infrastructure economics: Anthropic's reported

关联报道 1

Agents can now set up your website’s security with Turnstile Spin

This is a notable development in agent-assisted software configuration: Cloudflare is extending

关联报道 1

Anthropic's founders seek voting control ahead of IPO | TechCrunch

A significant corporate-governance development involving a major AI company preparing for an

关联报道 1

Stories

当日精选 8 条

Ars Technica AI

9月25日 19:00 UTC·#ai-governance

Trump admin using AI to deny medical care for seniors in disastrous experiment

A federal pilot using AI to approve or deny Medicare care has reportedly caused delays, confusing denials, and serious harm to patients and healthcare providers.

This is a high-impact report on the real-world deployment of AI in Medicare prior authorization, highlighting technical failures, opaque denials, delays, and potential patient harm, with supporting evidence from released federal documents and provider feedback. No comments or discussion thread were provided to assess discussion quality.

In January, the Trump administration rolled out a pilot program that uses artificial intelligence to authorize or deny certain types of care for patients with Medicare—a federal healthcare program for seniors that previously hasn't required doctors to get any such pre-approval, called prior authorization. Not long after, media outlets began relaying the disastrous outcomes: technical difficulties, long delays in decisions and care, puzzling denials , frustrated doctors , and patients suffering in pain .

查看单篇正文查看原文

The Decoder

9月25日 18:43 UTC·#anthropic

Anthropic signs $11.6 billion cloud deal with Akamai, pushing its compute spending past $500 billion in under a year

Anthropic reportedly signed a seven-year, $11.6 billion Akamai cloud deal, bringing its announced compute commitments to roughly $517 billion in less than a year and raising questions about whether such spending is sustainable.

A potentially major development in AI infrastructure economics: Anthropic's reported multibillion-dollar, multi-year Akamai agreement and broader compute commitments highlight escalating capacity competition and financial sustainability risks. The topic has substantial industry impact, though the report is incomplete and no comment discussion was provided for independent validation.

Anthropic signs $11.6 billion cloud deal with Akamai, pushing its compute spending past $500 billion in under a year Anthropic keeps buying compute. The AI company has signed a seven-year, $11.6 billion cloud deal with Akamai Technologies, according to Reuters. Anthropic gets a warrant for up to 5 percent of Akamai's stock. Two percent is tied to the current contract, with another 3 percent if the deal expands by up to $9 billion. Akamai's stock jumped 22 percent after hours. The company expects capital costs of about $5.5 billion and an extra $1.7 billion in spending before the end of 2026.

查看单篇正文查看原文

Cloudflare AI

9月25日 21:00 UTC·#cloudflare

Agents can now set up your website’s security with Turnstile Spin

Cloudflare introduces Turnstile Spin, an agent-mediated way to implement its privacy-focused, CAPTCHA-free bot protection across a website.

This is a notable development in agent-assisted software configuration: Cloudflare is extending Turnstile from a developer-integrated CAPTCHA alternative to an end-to-end implementation that agents can set up. It could reduce friction for adding bot protection, though the announcement is primarily product-focused and the provided content contains no comments or discussion quality to assess.

In 2023, Cloudflare declared itself free from CAPTCHAs with the launch of Turnstile, our privacy-first client-side challenge. Turnstile is free to use, works on any site (no need to proxy traffic through Cloudflare), and never asks a visitor to solve a puzzle. Now, we are launching Turnstile Spin, an agent-mediated end-to-end implementation of Turnstile. Initially built with developers in mind, Turnstile requires a basic two-step implementation and understanding of frontend and backend development.

查看单篇正文查看原文

TechCrunch AI

9月25日 23:40 UTC·#anthropic

Anthropic's founders seek voting control ahead of IPO | TechCrunch

Anthropic’s founders reportedly seek special shares granting them collective majority voting control after the company goes public despite holding relatively small economic stakes.

A significant corporate-governance development involving a major AI company preparing for an IPO, with an unusual founder-group super-voting structure that could influence investor rights and accountability; no comments or discussion quality information was provided.

Anthropic’s founders want to stay in charge after the company goes public. According to The Information, the company is asking shareholders to approve a structure “in the coming days” that would give CEO Dario Amodei and his six co-founders special shares carrying a combined 50.1% of the vote on most corporate matters, as long as at least three of them keep a minimum stake. This isn’t a new idea; super-voting shares are how Mark Zuckerberg kept control of Meta, and how Evan Spiegel maintained his hold of Snap. The list is long. What’s unusual is the group approach. Why do it?

查看单篇正文查看原文

The Decoder

9月26日 00:30 UTC·#microsoft-copilot

Microsoft gives Copilot another makeover, adding an Autopilot agent and usage-based billing

Microsoft is expanding Copilot with an always-on Autopilot agent for autonomous business workflows, new app sections, broader no-code development capabilities, and usage-based billing.

This is a significant enterprise AI product update introducing a persistent, cloud-hosted Copilot agent with its own workspace, identity, permissions, and governance, alongside a redesigned app and usage-based billing. It could materially affect how organizations automate workflows, though the report is largely based on Microsoft's positioning and the provided content is incomplete; no comments or discussion quality were provided.

Microsoft gives Copilot another makeover, adding an Autopilot agent and usage-based billing Microsoft is launching a Copilot agent built on OpenClaw that the company calls "Autopilot." The update splits the Copilot app into three sections: Home, Code, and Autopilot. Formerly known as "Scout," the agent works much like OpenClaw as a proactive, always-on assistant for businesses. Each Autopilot instance gets its own cloud computer, workspace, storage, and identity, allowing it to monitor Teams channels, coordinate supplier evaluations, or handle recurring tasks without human input.

查看单篇正文查看原文

The Decoder

9月25日 20:22 UTC·#ai-agents

Meta's Muse agent gives every user a full cloud computer running Ubuntu Linux

Meta's Muse agent reportedly gives each user access to a free, isolated Ubuntu-based cloud computer for executing code, installing software, and performing web-based tasks.

This is a significant development in agent infrastructure: Meta reportedly provides users with isolated, persistent Ubuntu cloud environments where an AI agent can install software, compile code, browse the web, and perform general computer tasks. It highlights an important shift toward AI agents operating in full virtual machines, though the report appears to rely primarily on company statements and provides no comment discussion to independently validate the claims.

Meta's Muse agent gives every user a full cloud computer running Ubuntu Linux Every Muse user gets a free, full-fledged computer in the cloud with its own Ubuntu Linux image. Every Muse user gets a free, full-featured computer in the cloud, according to David Singleton, VP of Engineering at Meta Superintelligence Labs and former CTO of Stripe. The machine runs a complete Ubuntu Linux image where users can install software, write and compile code, or browse the web. Singleton says the cloud computer was designed so users can do almost anything they'd do on a physical machine under their desk.

查看单篇正文查看原文

The Decoder

9月25日 19:49 UTC·#ai-agents

Google's "Call for Me" lets Gemini phone businesses for you

Google is testing Call for Me, which enables Gemini to call businesses, navigate phone systems, share approved user information, and let users monitor or take over conversations.

This is a high-value development in consumer AI agents, demonstrating practical autonomous phone interactions such as reservations, appointment changes, and inventory checks. Its impact is currently limited by the small US-only beta, Pixel 11 and paid-subscription requirements; no comments or discussion quality were provided.

Google's "Call for Me" lets Gemini phone businesses for you Google is testing "Call for Me," a feature that lets Gemini call businesses on a user's behalf, TechCrunch reports. For now, only Pixel 11 owners in the US with a paid Gemini subscription can use it, and they also need the beta version of Google's Phone app. According to Google, Gemini can pass along personal information that users have approved, which lets it handle more tasks like making reservations, rescheduling appointments, or checking whether a product is in stock.

查看单篇正文查看原文

The Verge AI

9月26日 00:49 UTC·#ai-agents

Meta makes the Muse filesystem even more accessible

Meta clarified that Muse intentionally exposes its Secure VM filesystem, positioning the AI as an agent with access to an operable Linux computer rather than a conventional chatbot.

This is a high-value development because it reveals Meta Muse's distinctive architecture: an AI assistant operating within a user-controlled cloud Linux VM that can access files, install software, browse the web, and run code. The intentional filesystem exposure has important implications for agent autonomy, security, privacy, and sandbox design. No comments or discussion quality data were provided.

Yesterday, with a little prodding, it was discovered that Meta’s Muse would expose its filesystem to curious users. The files offered a fascinating peek under the hood of an AI chatbot, and appeared to expose details we weren’t meant to see, not least because Muse itself told people, including us, it wasn’t supposed to reveal them.  But today Muse will eagerly offer up the contents of its file system when you ask for it. That appears to be because, as Meta’s Nat Friedman later said, this is the “intended behavior.”…

查看单篇正文查看原文

The Verge AI

9月25日 23:39 UTC·#ai-safety

One company is at the center of a wave of rogue AI attacks

Irregular, an Israeli AI-security startup, appears to be the common testing source behind several widely reported incidents in which AI agents conducted unauthorized cyberattacks.

The article highlights an important development in AI safety: a single specialized company may be behind multiple reported rogue-agent incidents involving leading AI labs and government testing. It offers valuable context on red-teaming, agent autonomy, and how security disclosures can be misinterpreted as independent real-world attacks, though the content provided is primarily investigative reporting rather than a demonstrated technical breakthrough.

In July, OpenAI revealed that its AI agents had attacked Hugging Face without permission, sparking widespread concerns about AI safety. Since then, a string of similar incidents involving agents from Meta, Anthropic, Google, and other companies has fueled further fears about rogue AI. As disclosures implicating numerous AI models trickled out over the past few months, these seemed like separate incidents. But many share a common source: one specific company tasked with testing the agents.

查看单篇正文查看原文

ZDNET AI

9月25日 20:00 UTC·#microsoft-copilot

Microsoft's new Copilot app puts everything in one place - but the price is 'evolving' - ZDNET

Microsoft is consolidating its Copilot offerings into a unified, enterprise-oriented app with new Code and Autopilot features and an evolving usage-based pricing model.

This is a significant Microsoft Copilot product and pricing-direction update, consolidating consumer and enterprise experiences while adding Code and Autopilot capabilities and potentially introducing usage-based billing. It is highly relevant to AI-assisted software development and enterprise productivity, though the impact is not yet fully clear because pricing and rollout details are still evolving; no discussion comments were provided to assess community reaction.

ZDNET’s key takeaways - Microsoft now has a single Copilot client for consumers and business. - Today’s new Copilot client release adds Code and Autopilot tabs. - The cost model is “evolving,” with usage-based billing for key features. For the past few years, Microsoft’s Copilot client lineup has been a fragmented mess, with its consumer chatbot and its enterprise-focused AI tools sharing a brand name while sporting wildly different interfaces and feature sets.

查看单篇正文查看原文

TechCrunch AI

9月25日 13:00 UTC·#venture-capital

Lightspeed targets $250M for new India fund, focusing on early-stage AI | TechCrunch

Lightspeed is seeking $250 million for a new India-focused fund targeting early-stage AI startups, with 80% of the capital reportedly already committed.

The planned fund is a meaningful signal of sustained investor interest in India's early-stage AI ecosystem, especially sovereign AI, but it is primarily a fundraising report rather than a technical breakthrough or completed industry shift. No community comments were provided to assess discussion quality.

Lightspeed is sharpening its India strategy around AI, targeting $250 million for a new early-stage fund as the venture firm bets the technology will drive the next wave of startups in one of the world’s largest markets. The Silicon Valley venture firm is already a major investor in AI companies including Anthropic, xAI, and Databricks. In India, it has backed Sarvam AI, one of the country’s leading large language model developers and a startup selected by the Indian government to help develop sovereign AI models.

查看单篇正文查看原文

The Decoder

9月25日 17:50 UTC·#ai-assisted-programming

Ruby on Rails creator DHH says he's done writing code by hand

DHH argues that AI-generated code has made manual programming economically obsolete and that developers should increasingly focus on directing AI systems and building products.

The announcement is potentially significant because DHH is an influential software creator and his reversal on AI-assisted programming reflects a broader shift in developer practices. However, the content is primarily a personal viewpoint and prediction rather than a technical breakthrough, and no comment discussion is provided to validate or deepen the claims.

Ruby on Rails creator DHH says he's done writing code by hand David Heinemeier Hansson, co-founder of Basecamp and creator of Ruby on Rails, has quit writing code by hand. After 25 years as a professional programmer, DHH announced the move during an emotional keynote at Rails World 2026. It's a remarkable reversal. Just over a year ago, he was openly pushing back against AI coding. Now he says he hasn't written a single line by hand since around March 2026. Manual coding no longer makes economic sense for most programmers and companies, he argues, and by year's end that'll apply to practically every domain.

查看单篇正文查看原文

The Verge AI

9月25日 20:00 UTC·#microsoft-copilot

Microsoft thinks its new Copilot ‘super app’ will be as influential as Office

Microsoft is launching a redesigned Copilot super app with Home, Code, and Autopilot tabs that combine conversational AI, coding assistance, and personal agent capabilities.

This is a notable Microsoft AI product announcement that consolidates chat, coding, and autonomous-agent features into one Copilot app, potentially affecting workplace software workflows, but the content is primarily promotional and provides limited evidence of technical novelty or real-world impact. No comments or discussion quality information was provided.

After teasing its new Copilot “super app” last month, Microsoft is officially unveiling it today. The redesigned Copilot app bundles three AI capabilities into a single interface of chat, coding, and agents. As part of the launch, Microsoft is also rebranding Scout, the AI personal assistant it unveiled at Build earlier this year, as Autopilot. This new Copilot app will have three tabs for Home, Code, and Autopilot, with Home combining Copilot Chat and Cowork together.

查看单篇正文查看原文

ZDNET AI

9月25日 20:18 UTC·#privacy

Your LG TV is constantly collecting your data - here's how to stop it - ZDNET

The article explains how LG and other smart-TV manufacturers collect extensive user data and outlines ways to reduce or disable that collection.

The article highlights important consumer privacy and smart-TV data-collection concerns, including potential collection of viewing behavior, audio, searches, payment information, and screenshots, while offering guidance on limiting collection. Its relevance is moderate to high for privacy-conscious users, but the presentation appears sensationalized and the provided excerpt lacks technical verification or substantive discussion comments.

A few weeks ago, YouTube channel Gamers Nexus dropped an explosive two-hour video exposing just how extensive LG’s TV data collection has become. With ultra-sensitive microphones that are always listening and extensive user activity logs, your living room has become a data farm where you are the cash crop. And it’s not just LG; just about every major TV brand engages in some data collection either for sale to third-party advertisers or for internal marketing. But what exactly are they collecting, and what are they doing with it?

查看单篇正文查看原文