Short News Archive

Jun 8, 2026

Only 26 percent of companies have full visibility into their AI costs, a KPMG survey finds.

The shift to token-based billing for AI services is putting finance departments in a tough spot, the Wall Street Journal reports. According to a yet-unpublished KPMG survey, just 26 percent of companies have full visibility into their AI spending. Half have limited oversight, and 22 percent have no transparency at all - or only find out what they've used after the bill arrives.

"It's a new resource that needs to be managed that didn't exist quite that way, and we're seeing exponential growth," Steve Chase, KPMG's global AI lead, told the WSJ.

KPMG is already working with several companies that burned through their annual token and cloud budgets within just a few months. One client saw a sixfold spike in token usage. Gil Luria, head of technology research at D.A. Davidson, expects the problem to hit more companies this year: "A lot of CFOs are going to see their Anthropic bill and freak out this quarter."

Analysts and executives the WSJ spoke with are drawing parallels to the pandemic-era cloud boom. Back then, companies poured money into cloud infrastructure, only to slash spending shortly after.

We took a closer look at why tokens are becoming a key business metric and how companies can prepare in our latest Frontier Radar #3.

Comment Source: WSJ

Jonathan Kemper

Jun 8, 2026

Short News

Moonshot AI, the Chinese company behind the Kimi chatbot, is looking for a valuation of up to $30 billion in a new funding round. That would be more than six times the $4.3 billion it was valued at in late 2025 and about 50 percent above the $20 billion from its previous round in May, the South China Morning Post reports, citing a person familiar with the matter. The Beijing-based company aims to raise between one and two billion dollars. Annual recurring revenue doubled to roughly $200 million by April.

On the technical side, Moonshot can hold its own against US providers. Its current flagship, Kimi K2.6, reportedly matches GPT-5.4 and Claude Opus 4.6 on coding benchmarks. But China's AI market is crowded. Moonshot competes with DeepSeek, Alibaba, and MiniMax. As recently as April, DeepSeek released V4-Pro, the largest open-weights model to date, bigger than Kimi K2.6 and far cheaper than the competition.

Moonshot is also preparing a possible IPO in Hong Kong and is reportedly unwinding its offshore corporate structure to make that happen.

Comment Source: SCMP

Matthias Bastian

Jun 7, 2026

Short News

"Chat is dead," a senior OpenAI employee told the Financial Times. The company is shifting from chatbots that answer questions to agents that handle tasks on their own. That shift is driving ChatGPT's biggest overhaul since its 2022 launch, turning the chatbot into a "superapp" bundling coding tools, AI agents, and partner integrations with companies like Canva and Booking.

Chief product officer Thibault Sottiaux told the FT: "It will transcend the actual surface . . . what we're building towards is where you have your own personal agent that is capable of helping you . . . across everything in your life, be it personally or at work." The reporting draws on more than a dozen current and former employees. OpenAI already confirmed it's working on a "super app."

In the coming weeks, OpenAI will redesign ChatGPT's web and mobile interface, adding features that steer users toward coding, image generation, and partner apps. Over time, those nudges should fade as the models learn to figure out what you need on their own. ChatGPT, Codex, and other product teams have already been merged under Sottiaux.

Comment Source: Financial Times

Matthias Bastian

Jun 6, 2026

Short News

Meta is developing a paid AI agent product called "Hatch" that could cost up to $200 per month. Hatch is designed to be a user-friendly version of the open-source tool OpenClaw and will handle tasks like creating software tools, scheduling appointments, and sending emails. Users describe what they need in simple language, and Hatch builds a working tool from that description. Microsoft with Scout and Google with Gemini Spark have recently introduced similar systems.

Internal documents show a free version and a "Hatch Plus" subscription with five to ten times higher usage limits. This puts Meta in direct competition with OpenAI and Anthropic, which charge $100 to $200 monthly for their top-tier subscriptions. A broader US launch is planned for July.

Hatch will also power Meta's planned AI hardware, including new smart glasses with a "supersensing" feature and an AI pendant set for internal testing in spring 2027. CEO Mark Zuckerberg sees AI agents as a way to open up new revenue streams beyond advertising, necessary to refinance Meta's massive AI infrastructure investments, which have already resulted in layoffs.

Comment Source: The Information

Matthias Bastian

Jun 6, 2026

Short News

Elon Musk's xAI spent months distilling Anthropic's Claude to train its own coding models, The Information reports. xAI trained its coding model directly on Claude's outputs. After Anthropic revoked official access in January, xAI engineers kept going through personal accounts and the intermediary service Blackbox AI. Musk previously admitted in court that xAI "partially" used OpenAI models to train Grok, calling it industry standard.

Internally, xAI looks troubled. The pretraining team shrank to under five people. Four Grok code leads left within months, along with many co-founders. One employee accidentally deleted critical training data, costing two to three weeks of work, according to The Information.

Musk's prediction from fall 2025 came true, just probably not the way he pictured it. | Image: via X

All that compute Musk bought up? He's now renting it to Anthropic via SpaceX and to Google instead of training his own models. Supposedly, it's just a stopgap.

Comment Source: The Information

Matthias Bastian

Jun 6, 2026

Short News

SpaceX signed a $920 million per month contract with Google, according to an SEC filing. The deal runs from October 2026 through June 2029, potentially netting SpaceX around $30 billion total. Google gets access to about 110,000 Nvidia AI chips to meet customer demand. A Google Cloud spokesperson told the New York Times it was a "short-term, timely agreement to ensure we have bridge capacity" for its Gemini Enterprise agent platform.

SpaceX is planning an IPO next week at a potential valuation above $1.7 trillion. Google holds about five percent of SpaceX, giving it a direct stake in a strong debut. The deal also doubles as advertising for Google's AI products, since SpaceX's IPO is drawing massive attention.

SpaceX previously locked in a $1.25 billion monthly deal with Anthropic. Both agreements position SpaceX as an AI infrastructure provider. Musk originally built the capacity for his own AI lab, xAI, which has lagged behind competitors. Leasing it out makes financial sense.

Comment Source: New York Times

Matthias Bastian

Jun 5, 2026

Short News

Florida is the first US state to sue OpenAI and CEO Sam Altman personally. The 83-page complaint accuses the company of marketing ChatGPT as safe while the chatbot delivers dangerous content to minors, facilitates violence, and drives users toward dependency. Attorney General James Uthmeier says OpenAI "put children at great risk" and threatens penalties in the billions. Numerous cases are documented where people were harmed by ChatGPT and similar systems.

The suit treats ChatGPT as a product subject to liability and a "public nuisance," an unusual legal approach that could set a precedent for chatbot regulation. The free version has no real age verification, the complaint says, even though tens of thousands of users are under 13. Data collection starts before users agree to the terms of service. The suit even argues AI use causes cognitive erosion.

The lawsuit doesn't mince words. | Image: Screenshot

The complaint cites internal allegations too: Altman allegedly cut short safety testing for GPT-4o, and OpenAI put just 1 to 2 percent of its computing power toward AI safety instead of the promised 20 percent. OpenAI has not commented.

Comment Source: My Florida Legal

Matthias Bastian

Jun 5, 2026

Short News

Microsoft CEO Satya Nadella has sharply criticized an internal memo proposing to make users "addicted" to the company's new AI agent Scout. That's "absolutely not a goal," Nadella wrote to about 50 top engineers, according to The Information. AI should empower people and create real value.

Not sure what this document is or who is writing and leaking this nonsense! They may want to go work elsewhere.
Satya Nadella, Microsoft

The memo came from Corporate Vice President Omar Shahine and laid out a three-phase plan from addictive app to agentic platform, as 404 Media first reported. Scout is built on the open-source software OpenClaw and was unveiled at Microsoft's Build conference.

Microsoft spokesperson Frank Shaw told The Information that Scout should help people get tasks done more effectively and ultimately lead to less screen time. Users should have choice and control over how they interact with the agent. Nadella's rebuke comes as social media platforms face growing criticism for using addictive design patterns to squeeze more usage time out of people.

Comment Source: 404 Media | The Information

Matthias Bastian

Jun 5, 2026

Short News

Microsoft partly trained its new MAI models on unlicensed web data. The technical paper shows Microsoft used Common Crawl, among other sources, as Simon Willison noted. Microsoft had previously claimed the MAI models were trained only on "enterprise grade, clean and commercially licensed data."

Microsoft's data pipeline for AI training: the tech giant also taps the open internet. | Image: Microsoft

Like other AI companies scraping the web, Microsoft is likely relying on fair use. The paper describes the data as a "mixture of publicly available and licensed human-generated data." For web data, Microsoft says it uses "a proprietary crawler that respects the Robots Exclusion Protocol (robots.txt) and related meta-tag and HTML controls, enabling site owners to manage how content on their sites is accessed and used."

That puts the burden of protecting content on site owners, like assuming anyone who doesn't lock their door consents to a break-in. Fair use remains contested, and courts are still sorting it out. In short, Microsoft does what every other AI company does, yet sells its training data as especially "clean." It isn't.

Comment Source: Microsoft

Matthias Bastian

Jun 5, 2026

Short News

The US National Security Agency is using Anthropic's Mythos AI model for offensive cyber operations, the Financial Times reports. Anthropic has placed about half a dozen engineers directly at the NSA to adapt the model and support its use. Mythos could be used to break into networks in China or Iran. Whether Anthropic's engineers are involved in active operations isn't clear.

The move comes while Anthropic is still in a legal fight with the Pentagon. The Department of Defense classified the company as a "supply chain risk" and tried to cut it from contracts because Anthropic wanted to restrict the use of its Claude models for mass surveillance and autonomous drones. The NSA delivery was exempt from that ban. Anthropic likely doesn't see a conflict with its own values: in its demands to the Pentagon, the company always pointed to protecting US citizens from AI threats.

Anthropic recently expanded access to Mythos to 150 organizations across 15 countries. President Trump also signed an executive order for voluntary safety testing of new AI models, which Anthropic viewed favorably.

Comment Source: Financial Times