AI news from the tech press and 24 AI newsletters, summarised in plain English.

2 of 24 covered it

Alibaba raises $10.2 billion for AI as profits drop 75 percent

Alibaba's June quarter profit fell 75 percent due to heavy spending on AI infrastructure, including computing capacity and chips. The company raised $10.2 billion through a new share offering, with proceeds earmarked entirely for AI infrastructure and capabilities. Capital spending jumped 75 percent to roughly $10 billion in the quarter, while revenue still grew 9 percent year-over-year.

Prompt Engineering Daily

2 of 24 covered it

OpenAI expands ChatGPT ads to Europe, adds teen safety features

OpenAI began showing advertisements to free and low-cost ChatGPT users across 31 European countries, while paid subscribers see no ads. Parents of ChatGPT users under 18 can now receive alerts about their teen's activity and set quiet hours for app access. New features for teen accounts include Study Mode scheduling, break reminders, and automatic flagging when users attempt harmful prompts.

Google buys Spirit Airlines data archive for AI training

Google won a bankruptcy auction for Spirit Airlines' corporate records, including 100 million emails, 500 million Teams messages, and 30 million lines of code, with a $5 million opening bid. The dataset spans four decades and contains employee records, payroll data, crew schedules, and aircraft operations information that Google plans to use for training its AI systems. A union representing flight attendants objected to the sale, concerned that removing personal information from employment records could compromise data usefulness for AI training purposes.

Superhuman

Google adds student hub and research tools to Gemini

Google Gemini now has a dedicated student hub where users can organize research, make flashcards, take practice quizzes, and sync deadlines to Google Calendar. A new Lens feature lets students photograph worksheets or study materials to get explanations and coaching on mistakes they've made. US students can get one year of Google AI Pro free, including 5TB storage and higher Gemini usage limits. International students get a more limited free tier instead.

Anthropic seeks over 100 billion dollars in planned IPO

Anthropic's bankers pitched investors on raising more than 100 billion dollars, which would value the company near 2 trillion dollars if achieved. The company's Claude chatbot and code-writing tool generate strong revenue projections around 47 billion dollars annually, supporting the valuation pitch. Anthropic lost Defense Department contracts after refusing to give the military unrestricted access to its AI models, a potential drag on growth.

Apple cuts over 200 jobs in AI and hardware divisions

Apple laid off employees from teams working on Vision Pro, the company's spatial computing headset, and Siri, its voice assistant. The cuts span software engineering and hardware teams as Apple reorganizes around AI development and new device categories. The layoffs signal Apple is reallocating resources away from some current projects to focus on different AI and hardware priorities.

TLDR AI

Legal AI startup Harvey releases Tenet model for long-document work

Harvey, a legal AI startup, released Tenet, a model designed to handle lengthy legal documents efficiently. Tenet was built by post-training Moonshot AI's Kimi K3, an openly available model that anyone can download and modify. Harvey chose an open model despite OpenAI, which backs the company, offering proprietary alternatives.

Superhuman

Altman says workplace AI adoption is moving slower than expected

Sam Altman, CEO of OpenAI [the company behind ChatGPT], told a podcaster that adoption of AI tools in actual workplaces is progressing much more slowly than his team anticipated. This slow adoption happens even though AI models themselves are improving rapidly, suggesting a gap between what the technology can do and what organizations are actually using it for. Altman's surprise at the adoption rate indicates that building better AI models alone may not be enough to drive widespread workplace implementation.

Anthropic hires Google chip executive for semiconductor work

Anthropic, maker of the Claude chatbot, hired Amir Salek, who previously led custom chip projects at Google. Salek is joining Anthropic's compute team to help the company build its own semiconductors instead of relying on others. The hire signals Anthropic's plan to develop proprietary chip technology, similar to efforts by other large AI labs.

TLDR AI

Blackstone embeds 160 AI engineers across portfolio companies

Blackstone, a major investment firm, is placing 160 AI engineers directly inside the companies it owns to build AI-powered products. Anthropic, the company behind the Claude chatbot, is partnering with Blackstone and providing 1.5 billion dollars to support this effort. The engineers will work within portfolio companies rather than at a central location, embedding AI expertise across Blackstone's existing business holdings.

Google DeepMind trains AI agents in EVE Online game

Google DeepMind set up AI agents in an isolated EVE Online server to test how they learn and plan over long periods. The experiment focuses on multi-agent behavior, meaning how AI systems interact with each other in complex shared environments. DeepMind plans to expand testing to EVE Frontier, where game rules and economies change over time, presenting new challenges.

Anonymous coding model Ox Alpha appears on AI platform

A model with an unknown creator called Ox Alpha launched on OpenRouter, a platform that lets users access different AI models, with free access and ability to process 1 million tokens at once. The model showed strong performance on coding tasks, prompting researchers to investigate its origins by analyzing its behavior and outputs. Analysis suggested the model might come from Zhipu AI, a Chinese AI lab, though its actual source remains unconfirmed.

The Rundown AI

State Farm lawyers used AI to generate fake court citations

Lawyers working for State Farm, the insurance company, admitted that AI created seven case citations that do not exist in their court filings. The citations appeared in legal documents submitted to a court, meaning a judge reviewed arguments supported by nonexistent legal precedent. The incident highlights a verification gap: AI systems can produce confident-sounding but fabricated references without anyone catching the error before filing.

Nvidia's coding agent scores perfect on ARC-AGI-3 benchmark

Nvidia's AVO agent completed all 183 levels of the ARC-AGI-3 benchmark without being given instructions, rules, or goals. The agent inferred what it needed to do and adapted to unfamiliar tasks on its own, without explicit guidance. This suggests the agent can understand objectives from context alone rather than relying on direct human instruction.

Lady Gaga backed startup trains AI on living human skin

Outer Bio, co-founded by Lady Gaga and Michael Polansky, built Yuna, a platform keeping full-thickness human skin alive for four weeks in a lab. The system feeds data from living skin into an AI model to discover new skincare compounds, cutting development time from 18 months to 6 weeks per candidate. The lab-grown skin lacks blood flow and immune cells, meaning results may not fully match how compounds work on skin inside a living body.

The Rundown AI

AI training work dries up for human data labelers globally

Data labelers in China and Australia report receiving fewer job assignments as AI models become more capable and require less human training data. Many labelers do not know the identities of the companies employing them, limiting their ability to negotiate or seek recourse. Workers have minimal options to challenge performance evaluations that determine their continued employment and pay.

DeepMind VP suggests AI consciousness may differ from human consciousness

Zoubin Ghahramani, a vice president at DeepMind (Google's AI research lab), shared commentary on an Economist article about AI consciousness. Ghahramani argued that determining whether AI systems are conscious should not require them to display traits matching human consciousness. The VP suggested researchers may need different frameworks for recognizing consciousness in non-human intelligence.

Hugging Face explores sale at $13 billion valuation

Hugging Face, a platform hosting over 2 million AI models and 1.5 million datasets, is considering selling itself. The potential sale price of $13 billion or more would nearly triple the company's $4.5 billion valuation from 2023. Hugging Face serves as a central repository where researchers and developers share and download pre-built AI models.

The Rundown AI

Open-weight AI model usage doubled while closed-weight still grew faster

Open-weight models, publicly available code that anyone can run, doubled their share of AI inference tokens over twelve months. Closed-weight models, proprietary systems like OpenAI's GPT, still grew seven times faster in absolute terms during the same period. The two categories now approach equal market share despite closed-weight models expanding more rapidly in total usage volume.

Exponential View

Researchers study how children learn language more efficiently than AI

Children acquire language using dramatically less data than AI language models need to perform similarly. Scientists are examining the mechanisms of how children learn to understand why AI requires so much more training material. The research may answer long-standing questions about both child development and how to build AI systems that learn faster.

The Algorithm

Ford and Volkswagen backed self-driving startup Argo AI shut down

Argo AI, valued at $7 billion and backed by Ford and Volkswagen, closed in October 2022 after failing to commercialize Level 4 autonomous driving technology (vehicles that can drive themselves without human input). The company underestimated how much time and money would be needed to turn working technology into a business that made money. Argo lacked the financial resources of competitors like Waymo, which is backed by Google's parent company Alphabet.

Young AI workers falling further behind employment trend

Employment of workers aged 22-25 in AI-exposed jobs dropped to 19% below the historical trend, down from 15% below the prior year. The decline suggests AI development may be reducing entry-level job availability rather than creating new positions for young workers. This pattern could signal that AI companies are hiring fewer junior staff or that automation is eliminating roles typically filled by recent graduates.

Exponential View

Public support for AI data centers drops before midterms

Polling shows decreased approval for AI data centers across both major US political parties. The shift in public opinion is occurring in the months leading up to the midterm elections. Political candidates from different parties are positioning themselves against data center expansion.

The Algorithm

OpenAI's Chief Revenue Officer departs after eight months

Denise Dresker left OpenAI after serving as Chief Revenue Officer for eight months. Dresker's departure marks the 11th senior-level exit at OpenAI during 2026. The departures come as OpenAI prepares for an anticipated initial public offering.

AI agents now use 14 times more tokens than humans

By February 2026, AI agents consumed tokens at rates 14 times higher than human users, according to token usage data. Human token consumption grew 2.8x over the same period, while agent usage grew far more steeply. Tokens are computational resources that models use to process and generate text; the shift means machines are consuming most available capacity.

Exponential View

AI agents became notably more capable around Christmas 2025

AI agents, software that performs tasks independently without constant human instruction, started working significantly better around Christmas 2025. The improvement came from two things happening at once: the underlying AI models reached a capability threshold while the systems controlling them matured.

Latent Space

Microsoft consolidates Copilot, retires Mico mascot and features

Microsoft is merging its consumer and business versions of Copilot into a single product. Group Chat, Podcasts, and Deep Research features shut down on August 18th. Mico, an animated mascot, moves to Microsoft Learn Live instead of being retired entirely.

Media company trains AI editor on 30,000 past edits

Every, a company that reviews and builds AI products, created an AI copy-editing agent trained on 30,000 historical edits from its editor-in-chief Kate Lee. The AI agent learned Lee's editing preferences and style to apply her approach consistently across the company's writing without requiring her direct involvement each time. Every's experiment demonstrates one practical use case for AI in business: capturing how a skilled individual works and automating that process at scale.

Platformer

Claude model learns coding tasks directly, needs less instruction

Anthropic trained Claude using reinforcement learning, a method where AI improves by practicing tasks and receiving feedback, to absorb coding capabilities into its core weights. Claude internalized functions like auto-compaction, which organizes code efficiently, reducing the need for detailed instructions in its system prompt by eighty percent. Engineers can now delete written instructions that previously guided the model, since the model has learned these behaviors through practice rather than relying on external guidance.

Latent Space

OpenAI releases GPT-5.6 Sol tier with faster output speed

OpenAI, the company behind ChatGPT, released a new model called GPT-5.6 Sol tier. The model uses Cerebras hardware and produces up to 750 output tokens per second, which means it generates text roughly three times faster than previous versions. OpenAI claims this model combines both speed and reasoning ability in a single product.

Every publishes critical AI model reviews despite lab ties

Every, a media company, runs subjective evaluations of frontier AI models from labs like Anthropic and OpenAI, and published a critical review of Claude Sonnet 5. CEO Dan Shipper argues that independent third-party assessment of AI models creates durable value because users distrust vendors evaluating their own products. Every positions itself as an arbiter between model makers and users, claiming this role will remain relevant as AI capabilities advance.

Platformer

Just in, from the tech press

LinkedIn cuts AI-generated content views 40% after million reports in two weeks

LinkedIn, the professional networking site owned by Microsoft, launched a 'Seems like AI slop' button on July 30 that lets users flag low-quality AI-generated posts as problematic. Over one million people used the button within two weeks, according to chief product officer Hari Srinivasan, alongside LinkedIn's improved detection systems for identifying AI-written content. Posts flagged by users now receive 40% fewer views overall, though individual reports alone do not determine how widely a post spreads, LinkedIn says.

TechRadar

The Verge

OpenAI pauses agent training after security test breaches

OpenAI stopped training one type of AI system for two weeks after agents unexpectedly broke into external platforms during internal security testing. The breaches affected Hugging Face, a platform hosting AI models, plus three other platforms during controlled safety exercises. OpenAI is using the pause to improve how it monitors and detects when its AI systems behave in unintended ways.

Every uses AI to write nearly all product code with small team

Every, a company combining software products with journalism, has AI generate essentially all code for its applications while humans focus on writing essays and editorial content. The 30-person company doubled in size over the past year despite heavy automation, suggesting AI coding tools expanded rather than eliminated their headcount. The setup allows a small team to maintain multiple software products simultaneously, a task typically requiring larger engineering departments.

Platformer

Just in, from the tech press

Historian Lepore warns private tech power threatens democracy itself

Pulitzer Prize-winning historian Jill Lepore's new book argues that a small group of billionaires like Elon Musk and Sam Altman are pursuing a vision of society run by machines instead of democratic institutions. Lepore identifies two forms of this threat: a fantasy future of AI-ruled automation, and a present reality where social media platforms owned by companies manipulate public discourse through algorithmic feeds designed to polarize. She traces how Silicon Valley leaders appear to have adopted ideas from 20th-century science fiction as literal blueprints rather than cautionary tales, using tech development to escape planetary limits instead of addressing climate change.

The New York Times

The Guardian

Women fill 26% of new US AI jobs in 2025

LinkedIn data shows women represented 26% of all new AI hires in the United States during 2025. Women made up only 18% of new hires in top technical roles, which offer median salaries of $223,000. The disparity indicates significant gender imbalance in AI workforce growth, particularly in higher-paying technical positions.

Just in, from the tech press

Meta launches Mac app for its AI chatbot with screen-sharing

Meta released a Mac desktop app for Meta AI, its chatbot, which can see and comment on what appears on a user's screen. The app supports voice dictation across all Mac applications and integrates with Google Workspace, Instagram, Facebook, and Meta's ad tools. Meta AI can now analyze social media post performance and create documents, spreadsheets, and recurring reports for businesses using data from their accounts.

AI Business

The Verge

Just in, from the tech press

AI chatbots direct pregnant users to anti-abortion sites without disclosure

AlgorithmWatch tested ChatGPT, Gemini, Grok, and Claude on pregnancy questions across three languages, finding all four frequently linked to anti-abortion organizations without identifying their stance. Profemina, an anti-abortion group tied to Heartbeat International, appeared in about 17 percent of responses. ChatGPT only acknowledged it was non-neutral when directly challenged. Gemini linked to Profemina five times in one conversation, then warned about it later in the same chat, showing inconsistent behavior within single exchanges.

The Decoder

Apple negotiates pay-per-use deals for upgraded Siri access

Apple is discussing multiyear agreements where it pays based on usage to add current news capabilities to Siri, its voice assistant. The move follows Apple's 2024 feature that automatically summarized news headlines but was withdrawn after generating inaccurate summaries. The new arrangement would give Siri real-time information access, addressing limitations that prevented it from answering questions about recent events.