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

Anthropic deploys security scanner using latest Claude model

Anthropic released Claude Security, a tool that scans computer code for vulnerabilities and suggests fixes, now running on Claude Mythos 5, their most capable model. Enterprise customers can access the scanner in public beta. A human must approve every suggested patch before it takes effect. Security companies protecting hospitals, utilities, and banks are integrating Claude Mythos 5 into their products. End users see only the results, not the AI itself.

Meta launches Mac app for AI assistant with screen sharing

Meta released a Mac app for its AI chatbot that can see your screen, answer questions about it, and dictate text across any application. Businesses can now connect Instagram, Facebook, and Google Workspace accounts to Meta AI to get performance insights and automate tasks like creating proposals and weekly reports. The app lets Meta AI analyze what posts perform well, suggest content, and research competitors using public data to help merchants improve their marketing.

AI Breakfast

Apple Music will require AI content labels later this year

Music distributors must tag any track where AI generated a significant portion, making labels mandatory rather than voluntary starting later in 2024. Apple defines AI-generated music as content primarily derived from generative AI services like Suno and Beatoven, not human-created work with AI assistance. The policy follows Spotify's recent decision to label AI artists and exclude them from recommendations, plus Tidal's approach of not paying for AI-detected uploads.

The Rundown AI

xAI releases Grok Bot, an AI agent that automates app tasks

Grok Bot entered beta on August 11, 2026 as an AI agent capable of logging into applications, navigating screens, and completing tasks without requiring API connections. The agent works by interacting with apps the way a human would, clicking buttons and entering data rather than relying on direct data integration. Early users including prominent technologists shared demonstrations on social media, creating widespread visibility within hours of the beta launch.

Prompt Engineering Daily

Anthropic prepares IPO as public opposition to AI data centers grows

Anthropic held confidential investor meetings in June and expects to go public at a valuation around $2 trillion, potentially exceeding SpaceX's recent $85.7 billion offering. A Gallup survey found seven in ten Americans oppose building AI data centers in their area, with politicians from both parties now pushing back on development. Anthropic's CFO fielded investor questions about competition, profit margins under pressure from free open-source models, and risks if data center construction slows.

AI Breakfast

AI agents started working effectively around late 2025

Functional AI agents emerged in late 2025, representing a shift from experimental to practical systems. The capability jump resulted from both better underlying models and more mature software frameworks around them, not model improvements alone. Lukasz Kaiser, a researcher involved in the field, confirmed the timing of the shift but noted its exact causes remain unclear.

Latent Space

Reddit user reports Claude lost $31K in autonomous trading experiment

A Reddit user claimed they gave Claude, an AI chatbot made by Anthropic, access to a trading account for a month and it lost $31,000. The user posted the report as a cautionary example that autonomous AI agents can make poor financial decisions when given real money to manage. Community responses emphasized safer testing practices: use simulated trading first, limit position sizes, and set automatic stopping points before letting AI agents trade real funds.

DeepSeek Harness becomes fastest growing GitHub repository

DeepSeek Harness, a tool that lets developers use DeepSeek's AI model similarly to how Anthropic's Claude handles coding tasks, launched on GitHub. The repository grew faster than any other project in GitHub's history, suggesting rapid developer adoption and interest.

AI Breakfast

Anthropic releases Claude Code with autonomous system access

Claude Code, a new version of Anthropic's Claude chatbot, can now directly execute bash commands and access files without human approval. The release happened in February 2025 when reasoning models (systems trained to think through problems step by step) became reliable enough that developers felt safe removing safety restrictions. For years, Anthropic had intentionally limited what models could do to prevent errors. This is the first time they deliberately expanded those permissions.

Latent Space

AT&T routes simple AI tasks to cheaper open models

AT&T deployed a system that automatically sends routine work to inexpensive open-source AI models and only routes complex tasks to expensive systems like OpenAI and Anthropic. The routing approach reportedly reduced costs by 80-90 percent on some applications while maintaining acceptable quality, according to reporting. This reflects a shift where companies choose different AI models for different tasks instead of committing to single vendor platforms.

Mistral releases search tool that guides AI through documents

Mistral, the French AI company, released Agentic Search, which gives AI models five operations to navigate documents rather than accepting the first result. In internal tests on financial documents, the tool improved answer correctness from 26.7% to 86%, though Mistral conducted the measurements itself. The system works by having models inspect long documents and verify answers through a loop of checking, rather than stopping at the first retrieval.

Claude's system prompt shrunk 80% as model absorbed harness functions

Anthropic reduced Claude Code's system prompt by 80% after the model learned to handle tasks internally through reinforcement learning in real environments. Models are absorbing capabilities that previously required external instructions, like managing context window size and calling tools, into their learned weights. As models internalize these functions, engineers can remove the scaffolding code that once directed behavior, simplifying the overall system architecture.

Latent Space

Nvidia licenses Poolside AI coding startup for $6 billion

Nvidia paid $6 billion in a licensing deal to use Poolside's AI code-writing technology, which the startup developed to help programmers write software. Nvidia made a separate $1 billion investment in Poolside and offered jobs to 109 of its employees, though the founders stayed to run the company. The licensing agreement is non-exclusive, meaning Poolside can license its technology to other companies alongside the Nvidia deal.

PagedAttention brings virtual memory technique to AI model memory

PagedAttention applies virtual memory concepts, a computer architecture idea, to how AI models store information during processing. The KV cache stores key-value pairs that models need to track context, and it consumes substantial GPU memory when processing long texts. This technique lets attention kernels (the part computing which words matter most) access memory more efficiently without requiring redesign.

Anthropic launches Claude Academy with 355 learning resources

Anthropic, the company behind Claude chatbot, created Claude Academy as a central hub for learning how to use its products. The Academy contains 355 tutorials, prompting tips, and examples across Claude, Cowork, Code, Tag, and the API interface. Anthropic also released Concise Mode for Claude Code, a feature that makes the model keep responses shorter.

AI assistance plus training boosted Pakistani judge case resolution

A study of 1,559 Pakistani judges found that AI tools combined with targeted training increased the rate at which judges resolved cases by 6.3 percent. Case resolution improved without measurable drops in the quality of written decisions or increases in appeals of those decisions. The experiment was randomized, meaning some judges received the AI plus training while others did not, allowing researchers to isolate the effect.

Researcher proposes new methods to evaluate AI intelligence

Melanie Mitchell argues that AI systems think in ways fundamentally different from human reasoning, making standard evaluation approaches inadequate. Mitchell suggests borrowing assessment techniques from infant and animal psychology to better understand how AI actually processes information. She outlines six principles intended to improve how researchers measure and evaluate AI capabilities.

ChatGPT stopped citing Reddit content after July

ChatGPT's citations of Reddit dropped from nearly 4 percent in mid-July to 0.5 percent by August 14th, an 86 percent decline. Reddit was historically among the most-cited sources for AI chatbots before the sharp decrease. OpenAI has not publicly explained why the citations fell so dramatically.

Goldman Sachs: AI already reducing jobs in developed economies

Goldman Sachs research identified five sectors where AI is currently displacing workers: call centers, software publishing, consulting, advertising, and entry-level positions. The impact is most visible in developed economies rather than emerging ones, suggesting wealthier countries are adopting AI tools faster. Job losses are happening now, not in some distant future scenario economists often debate.

Anonymous provider releases Ox Alpha reasoning model for coding

Ox Alpha, a new reasoning model, is available through OpenRouter, a service that routes requests to various AI providers. The model is designed for coding tasks, agentic work (systems that act autonomously), and complex reasoning with both text and images. Its creator remains unnamed, distributed through a third-party platform rather than a known company.

Tech giants hiding 3 trillion in AI spending from investors

Wall Street Journal investigation found nine major tech companies are carrying 3 trillion dollars in AI commitments that don't appear on their official financial statements. These hidden expenses make it difficult for investors to understand the true financial obligations and risks these companies have taken on. Off-balance-sheet accounting allows companies to keep certain commitments out of public view, creating a gap between what appears in financial reports and actual spending.

AI system identifies two new patterns in breast cancer cells

Researchers at University of Northampton used CenSegNet, an AI platform, to analyze tumor samples and found two distinct patterns in centrosome abnormalities (structural defects in cells) that correlate with cancer progression. The patterns discovered appear to link centrosome abnormalities to patient outcomes, potentially offering new ways to predict or understand cancer behavior. The finding demonstrates AI's capacity to detect subtle patterns in medical images that might escape human observation or traditional analysis methods.

The Rundown AI

Startup trains smaller AI model to handle changing tool interfaces

TaoLive developed a training method called Harness-Aware Training that teaches a 35-billion-parameter model (a mid-sized AI system) to adapt to different tool interfaces instead of memorizing one fixed version. The model learns during training to interpret variable tool names, schemas (the structural blueprints of tools), and prompt structures, making it flexible across different setups. The company tested this approach specifically in live commerce, the real-time sales environment where product catalogs and interfaces frequently change.

Alibaba's Qwen model becomes most downloaded open-weight model

Alibaba's Qwen model family reached 3 billion downloads in six months, making it the most downloaded open-weight model available. Qwen surpassed models from Alphabet and Meta, despite those American companies being earlier and more prominent in open-weight AI development. Open-weight models are freely available AI systems anyone can download and run on their own computers, contrasting with closed services like ChatGPT.

Asana used AI to replace testing system in two weeks

Asana, a work management software company, deployed OpenAI's Codex, an AI system trained to understand and write code, to remove an obsolete testing framework. The project completed in two weeks at a cost of approximately twelve thousand dollars, compared to an original estimate of five years and six million dollars. The speed and cost reduction suggests AI coding tools can handle large modernization tasks that were previously considered prohibitively expensive or time-consuming.

The Rundown AI

Interactive guide maps five parallel training strategies for AI models

A new interactive guide explains data parallelism, FSDP, tensor parallelism, pipeline parallelism, and expert parallelism. These are different ways to split model training work across multiple computers. The guide shows how hardware capabilities and communication patterns between computers determine which strategy works best in different situations. Each strategy has performance limits. The guide helps identify which bottleneck (communication delay, memory limits, or compute limits) will hit first for a given setup.

Firefox adds optional AI chatbot to organize browser tabs

Firefox partnered with Exa, a search company, to build Smart Window, a chatbot users can enable or disable. Smart Window pulls current web information and sorts open browser tabs into organized groups automatically. Users choose which AI model powers the feature and control whether their data stays on their computer or goes to Exa's servers.

Media company builds AI editor trained on 30,000 past edits

Every, a media company focused on AI reviews and products, trained a copy-editing AI system by analyzing 30,000 historical edits from its editor-in-chief Kate Lee. The system allows the company to apply Lee's editing decisions and style across all its writing without requiring her direct involvement on each piece. Every sees this as a model for how companies might capture and scale individual employees' expertise through AI rather than hiring more staff.

Platformer

Google brings Antigravity coding agents to enterprise customers

Google added Antigravity, an AI agent that writes code, to Gemini Enterprise subscriptions for eligible customers. The tool now works inside four developer environments: VS Code, Visual Studio, JetBrains, and Zed, so programmers can access it where they already work. Enterprise administrators gained controls over sandboxing, tool permissions, spending limits, user identity, and audit logs for compliance and security.

LLM tool breaks after OpenAI library removes dependency

LLM, a command-line tool for running AI models locally, stopped working on fresh installations when OpenAI's Python library dropped its httpx dependency. LLM had been indirectly relying on httpx through the OpenAI library without declaring it as its own dependency, creating a hidden fragility. Version 0.32.1 of LLM fixes the issue by explicitly including httpx as a declared dependency.

Simon Willison

Newsletter CEO discusses balancing criticism with dependence on AI labs

Every, a newsletter covering AI, relies on early access to models from Anthropic and OpenAI to test products and write about them. Every's CEO Dan Shipper says the publication can publish critical reviews despite this dependence because AI companies cannot fairly evaluate their own products. Shipper argues Every's value comes from acting as a neutral third party that readers can trust more than the companies themselves.

Platformer

Anthropic adds computer control and browser access to Claude

Claude, Anthropic's AI chatbot, can now control computers and browse the web as part of a unified agent building system. Teams can upload procedures once and version them for reuse, rather than rebuilding instructions each time. The system reduces redundant browser interactions by letting agents reuse previously saved files and procedures.