AI news from the tech press and 24 AI newsletters, summarised in plain English.
Three stories a day, by email. We pick the three that actually mattered and send nothing else.
Ranked by how many of the 24 newsletters we read covered each story.Choose your own
OpenAI pauses training after models show security risks
OpenAI halted reinforcement learning training for two weeks and delayed its largest planned training run after finding its Astra model may develop cyberattack capabilities. The company implemented new safety measures including network isolation, stricter sandboxes, and a monitoring system that flags suspicious behavior within 30 minutes. OpenAI cited rapid progress in internal research and a security incident at Hugging Face, another AI company, as reasons for the slowdown.
The Rundown AI
Latent Space
The Algorithm
Anthropic's revenue run rate hits $65 billion in July
Anthropic, which makes the Claude chatbot, reported a $65 billion annualized revenue run rate by late July, up from $47 billion in May and $10 billion for all of 2025. Second quarter preliminary revenue reached $11.5 billion, a 14-fold increase from the same quarter last year, and the company achieved positive adjusted operating income. Anthropic is preparing for an IPO expected as soon as fall 2026 and is seeking a $2 trillion valuation, which would be the largest market debut on record.
Superhuman
Exponential View
Etched raises $700M at $21B valuation, doubling in one month
Etched, an AI chip startup, raised $700 million led by Jane Street, a trading firm that tested the hardware before investing. The valuation jumped from $10.3 billion in July to $21 billion now, gaining roughly $11 billion in a single month. Etched designed two custom components: a low-voltage prefill chip that processes input faster, and cluster-scale memory that connects multiple chips with shared fast memory.
TLDR AIThe Rundown AILatest articles
OpenAI launches ChatGPT version with protections for teenagers aged 13-17
OpenAI released ChatGPT for Teens on Tuesday with restrictions on self-harm, eating disorder, and sexual content discussions, plus blocks on the chatbot claiming to have emotions. The platform uses behavioral signals like login patterns to estimate user age and automatically route users under 18 to the teen version, without direct age verification. Parents can link accounts to set quiet hours, receive safety alerts for high-risk situations, and control Study Mode settings that guide homework help rather than providing answers.
The Rundown AI
MindstreamTwo open-source AI development tools released
Miles, a reinforcement learning framework developed with 72 contributors over nine months, became available for training language models like Kimi K3 and DeepSeek V4. Mojo, a programming language for GPU computing, released version 1.0 and open-sourced its compiler under Apache 2 license after shifting away from full Python compatibility. Both projects moved from closed to open development, though Mojo represents a strategic pivot from its original goal of being a complete Python alternative.
Latent Space
Simon WillisonNvidia invests $1.5B in OpenAI's Ohio data center project
Nvidia will invest $1.5 billion in SB Energy and provide up to $105 billion in financing for a data center near Cincinnati that will supply computing chips exclusively to OpenAI. The facility will start with 4.25 gigawatts of computing capacity in 2028, expandable to 8 gigawatts, built on land formerly used for uranium enrichment by the U.S. Department of Energy. A $33 billion natural gas power plant will be built to support the data center, and SoftBank and OpenAI are existing investors in SB Energy alongside Nvidia.
Prompt Engineering DailyAnthropic adds invisible watermarks to Claude text for EU compliance
Anthropic will embed undetectable patterns into Claude's text output to comply with an EU regulation requiring AI companies to mark synthetic content by December. The watermark works by slightly constraining random word choices at non-critical moments, like choosing 'grey' or 'overcast' to describe weather, leaving a pattern detectable only with a special key. Anthropic says watermarks will not degrade writing quality or increase costs, though some critics argue constraining word choice will force less precise language overall.
AI BreakfastAlibaba's Qwen3.8-27B becomes top locally runnable open model
Alibaba released Qwen3.8-27B, a model people can run on their own computers that ranked first among similar models in Cline, a coding tool, within four days. The model scores well on standard tests, but some developers noted these benchmark scores don't fully reflect how well it actually performs at real coding work. The release demonstrates increasing availability of powerful open models that don't require company servers and come without content restrictions.
Latent SpaceChinese firms access advanced Nvidia chips remotely via Southeast Asia
ByteDance and Tencent have obtained computing power from Nvidia's most advanced chips by renting access through data centers in Malaysia, Thailand, and other Southeast Asian countries. U.S. export controls ban shipping these chips directly to China, but do not restrict remote access to them, creating a legal loophole that Chinese AI companies are exploiting. New Chinese AI models from Moonshot, DeepSeek, and Alibaba have improved significantly in recent months, with access to overseas compute identified as a key factor in their advancement.
The AlgorithmAI leaders clash over regulation and market concentration
Anthropic CEO Dario Amodei proposes federal review of powerful AI models before release, arguing scaling laws inherently concentrate power regardless of regulation. Critics including investor Gavin Baker, former White House adviser David Sacks, and Meta researcher Yann LeCun say Amodei uses fear to justify regulatory advantage for his company. Amodei counters that open-source AI just shifts concentration to whoever owns the most computing power, and that well-designed regulation can restrain corporate power rather than entrench it.
AI BreakfastGoogle partners with AMD on next-generation AI chip design
Google and AMD are collaborating to build a 10th-generation TPU, Google's custom AI processor, with integrated CPU cores on the same physical chip. The design combines AMD's x86 processor technology with advanced 3D stacking techniques to reduce the distance between CPU and GPU-like components. The partnership targets workloads where software agents need to make quick decisions, reducing delays between computation steps that currently require separate chips to communicate.
TLDR AIMozilla adds optional AI features to Firefox browser
Firefox now includes AI-powered translation, tab organization, and a sidebar for accessing chatbots like Claude and ChatGPT. Users can disable all current and future AI features with a single setting rather than managing them individually. Mozilla treats AI as an optional addition to the browser instead of requiring it, differing from browsers built around AI from the start.
Prompt Engineering DailyAI-generated character gains 1M TikTok views in sorority recruitment experiment
Olivia Moore, an investor at a16z [venture capital firm], created Janie, a fictional 19-year-old character using AI video generation tools. Janie's sorority recruitment videos accumulated roughly 1M views on TikTok within a week, created with about $100 and 30 minutes of daily work. Moore later revealed the experiment was not real. Viewers responded positively overall, though the case illustrates how AI video can become difficult to distinguish from authentic footage.
The Rundown AICerebras claims faster AI performance than Nvidia's chips
Cerebras announced a new AI supercomputing system built on a single wafer of silicon instead of multiple separate chips. The company claims its design is faster and produces more text output per second than Nvidia's leading AI accelerators. Cerebras eliminated connection delays between chips by putting everything on one piece of silicon, reducing a traditional bottleneck.
TLDR AIUK employers report AI creating jobs, but benefits skew toward large firms
54% of UK employers say AI has led to job creation, with a quarter now hiring for AI skills roles. Larger companies with over £10 million turnover have two-thirds of required AI skills already in-house. Many employers are training existing staff rather than hiring new workers for AI positions from outside.
Prompt Engineering DailyLegal AI startup Harvey launches Harvey II with context memory
Harvey, a legal AI startup, released Harvey II, a system that remembers details about specific legal cases across conversations. The new version can learn and adapt to individual lawyers' writing styles and preferences within a single legal matter. Harvey also introduced its first proprietary AI model built in-house rather than relying on other companies' models.
The Rundown AINvidia reserves advanced chip production capacity through 2028
Nvidia secured manufacturing slots at TSMC for Feynman, its next AI chip architecture arriving in late 2028. The chips will use 1.6nm process technology, which refers to transistor size and represents a step forward in miniaturization. Feynman incorporates nanosheet transistors and backside power delivery, both techniques for fitting more computing power into less physical space.
TLDR AIZhipu's GLM-5.3 improves without adding more parameters
Zhipu, a Chinese AI lab, released GLM-5.3, an updated version of its large language model that performs better on benchmark tests while keeping the same 753 billion parameters, the numerical weights that define how a model works. The improvements came from better training methods after the initial model building, specifically using reinforcement learning (training by rewarding desired behaviors) and techniques to compress knowledge from larger systems into smaller ones. The result suggests AI labs may focus less on simply building larger models and more on refining how they train existing ones to handle complex reasoning tasks.
Latent SpaceAI math startup formalizes 246-year-old prime gap theorem
Axiom AI, a mathematics-focused AI startup, formalized the 246 theorem, which concerns gaps between prime numbers. Formalizing means translating a mathematical proof into a form a computer can verify as logically correct. This theorem had stood for 12 years without being formalized, and Axiom's system completed the task.
The Rundown AILiquid cooling monitoring detects AI hardware heat problems earlier
AI accelerators increasingly use liquid cooling systems, which can hide thermal problems until temperature alarms activate. Monitoring the entire cooling path, not just individual component temperatures, reveals thermal stress sooner. Better visibility into cooling system health helps prevent hardware failures and manage workload on control systems.
TLDR AINVIDIA tool cuts Hugging Face model deployment to two commands
NVIDIA released TensorRT Model Connect, which converts models from Hugging Face, a popular model repository, directly into optimized inference format without intermediate steps. Infrastructure teams can now deploy these converted models using C++ APIs with minimal setup, reducing complexity for engineers working with machine learning systems. NVIDIA built the tool partly using Codex, an AI code assistant, reflecting a broader shift toward teams using AI agents for infrastructure work.
Latent SpaceAndrew Yang proposes $15k annual payments for AI data use
Yang, a former U.S. presidential candidate, called for $15,000 yearly payments to families. The proposed payments would compensate people whose public data AI companies used to train models. Yang frames the payments as a way to distribute wealth generated by AI development.
The Rundown AIAI datacenters explore higher voltage power distribution systems
Datacenters are testing 800VDC power systems as an alternative to current 48V setups, which could reduce energy lost as heat during conversion from grid power to computer chips. The shift would require less copper wiring and special semiconductors called silicon carbide and gallium nitride to manage the higher voltage safely. Implementation across the industry could take years because it requires redesigning how power flows through entire datacenter facilities.
TLDR AICursor explains Git storage design for AI coding agents
Cursor, a code editor that uses AI assistants, published technical details on how it stores data in Git repositories to handle heavy automation workloads. The company framed Git hosting as essential infrastructure for AI agents rather than a standard development tool, due to the volume of automated code changes agents generate. The retrospective addressed specific engineering challenges that arise when AI systems frequently modify and commit code to repositories.
Latent SpaceStudy finds AI agents struggle with open-ended research tasks
Researchers tested AI agents on open-ended tasks requiring creativity and judgment, finding they cannot yet complete this type of work. The limitation raises questions about whether AI systems can improve themselves recursively, or if they can only advance through narrower, more defined improvements. The study suggests a potential ceiling on how far AI can self-improve without human guidance on more complex, unstructured problems.
The AlgorithmChina's AI hardware sales abroad show clearer demand than US investment
US AI investment often involves suppliers financing their own customers' purchases, making it hard to separate real demand from circular money flows. China is expanding sales of physical AI hardware and robots to other countries, creating verifiable demand signals through customs records and production data. Physical products shipped overseas provide clearer evidence of actual AI adoption than investment funding, which can recycle between related companies.
TLDR AIAI inference speed claims jump sharply across multiple vendors
Alibaba's Qwen model reportedly generated 70 tokens per second on Apple's M5 Max chip, showing faster processing on consumer devices. Cerebras announced a CS-4 system claiming 1000 tokens per second for large models, a significant jump in datacenter performance. Faster inference, meaning quicker AI response times, now affects user experience, operating costs, and how governments view AI competitiveness.
Latent SpaceFlock releases AI tool that identifies drivers by movement patterns
Flock, a company that sells surveillance tech to police, created an AI system that recognizes individual drivers based on how they move, not just license plates. The tool analyzes driving behavior patterns to match vehicles across multiple camera feeds, expanding police tracking capabilities beyond traditional identification methods. The technology raises questions about surveillance scope, though the newsletter presented both the company's rationale and privacy concerns without taking a position.
The AlgorithmUS moves to ban Chinese optical transceivers for AI networks
The FCC reportedly plans to restrict Chinese optical transceivers, components that connect AI systems and transfer data between computers. US officials cite concerns about data theft and reliance on Chinese suppliers for critical infrastructure. American manufacturers currently cannot produce optical transceivers at the scale China does, creating potential supply shortages if the ban takes effect.
TLDR AIMicrosoft consolidates Copilot apps, retires animated mascot
Microsoft merged its separate consumer and business Copilot applications into a single platform on August 18. The company removed several features including Group Chat, Podcasts, Deep Research, and Copilot Labs from the unified app. Mico, an animated mascot, was retired from the main Copilot experience and moved to Microsoft Learn Live, a separate educational platform.
MindstreamEgyptian developer builds AI app to help blind users see
A blind developer from Egypt created an application using cameras and AI to answer questions about what is around users. The app lets people with visual impairments understand their environment by describing objects and scenes in real time. The project shows how AI tools can be adapted to solve accessibility problems for people with disabilities.
The AlgorithmGroq, AI chip startup, reaches $3.5 billion valuation
Groq, which makes specialized processors for running AI models, achieved a $3.5 billion valuation in a new funding round. The company acquired intellectual property from Nvidia, the dominant chipmaker, as part of this funding. The valuation reflects investor confidence in Groq's ability to compete in the market for AI-specific hardware.
TLDR AINine tech firms carry $3 trillion in unreported AI spending deals
Major companies including Alphabet, Amazon, and Meta have committed to $3 trillion in AI costs not fully shown on financial statements, with $1.2 trillion in data center leases and $1.9 trillion in hardware orders. These hidden commitments are large enough to push some companies into negative free cash flow, meaning they are spending more money than they generate from operations. The deals are becoming larger and more complex, making it difficult for investors and regulators to see the full scale of each company's AI spending obligations.
SuperhumanScientist creates embryo-like structures without eggs or sperm
Jacob Hanna, a Palestinian stem-cell scientist, developed synthetic embryo models that mimic real embryos using neither sperm nor eggs nor fertilization. The synthetic models could help researchers understand how human embryos develop in their earliest stages and potentially advance regenerative medicine applications. The work raises ethical questions about how permissible this type of research should be and where boundaries ought to exist.
The AlgorithmEtched recruits senior hardware engineers from Nvidia
Etched, a startup building AI hardware, is hiring experienced engineers who previously worked at Nvidia, the dominant chip maker. The company is targeting senior-level positions including hardware engineers and system architects, roles that require years of specialized experience. This hiring pattern suggests Etched is moving beyond early stages and needs expertise to develop its own chip designs.
TLDR AINous Research adds Bot Mode to Hermes Desktop app
Bot Mode lets users create multiple AI agents within Hermes Desktop, each with different abilities and separate memory storage. Agents can share information with each other to work together on tasks across the application. Hermes Desktop now runs on macOS, Windows, and Linux.
SuperhumanTown launches AI assistant that organizes company information
Town, a new startup, released Townies, AI assistants that automatically build internal knowledge bases from employee emails and calendars. The system aims to let companies organize themselves with less manual effort, though Town's CEO estimates AI currently handles only 10-20 percent of knowledge worker tasks. The startup raised $55 million before its public launch in June and faces questions about how it handles employee privacy.
PlatformerEuropean Central Bank warns continent losing competitive edge to U.S., China
Christine Lagarde, head of the European Central Bank, said Wednesday that Europe's three-pillar post-war growth model is cracking: global trade is shrinking, cheap energy access is gone, and U.S. military leadership is withdrawing. Trump's tariffs on EU goods (initially 20%, then reduced to 15%) and threats to reduce U.S. security commitments in Europe are forcing companies to prioritize resilience over efficiency, reducing investment and economic output. Europe risks repeating its dotcom mistake by missing the AI revolution. U.S. tech giants are worth $23 trillion combined versus €1.37 trillion for Europe's top 34 tech firms, with unclear whether European AI investment will scale continent-wide.
CNBCCursor launches code-hosting platform to rival GitHub amid outages
Cursor, an AI startup now owned by SpaceX, released Origin this week, a code-hosting platform that lets developers collaborate on projects, manage edits, and store code repositories like GitHub does. Origin works alongside GitHub rather than replacing it. Developers can sync their existing GitHub repositories into Origin and move code between the two platforms. Cursor plans to add agent-native features (AI tools that can take actions autonomously) to Origin soon and is building an ecosystem of apps for broader coding work.
TechCrunch
VentureBeatOpenAI and Anthropic add learning-by-example features to chatbots
OpenAI launched Record & Replay and Anthropic launched Record a Skill, both letting their AI systems learn tasks by watching a user perform them once instead of reading written instructions. Demonstrating a task captures unspoken details that written prompts miss: a person asking their manager to review expensive meals, or treating client dinners differently from team lunches. Researchers call this tacit knowledge, the context people understand but cannot fully explain. One study estimated 40% of company knowledge exists only in employees' heads, never written down.
Fast CompanyAnthropic's Claude runs protein design experiments, beating typical success rates
Anthropic tested Claude models (Mythos Preview and Opus 4.8) on designing minibinders, small proteins that block target proteins, a foundation for drug development. Of 1,320 designs Claude created against 15 protein targets, 354 actually bound in lab tests, a 26.8 percent success rate versus the typical 10 to 15 percent in the field. Claude did not build new protein prediction software. It installed and orchestrated existing open-source tools through a 16,000-word instruction set, automating what usually requires human expert decisions and days of work.
The DecoderAmazon destroys rare books at Las Vegas facility to train AI models
Amazon is buying rare and out-of-print books in bulk, physically destroying them by cutting spines and scanning pages at a Las Vegas warehouse called VGT3 for use in training its AI models. Rare books are valuable training material because they contain text predating 2022, avoiding the risk of AI models degrading from ingesting too much AI-generated content, a problem called model collapse. Amazon workers documented in forums that the facility ran critically low on books to scan earlier this year, suggesting the company is systematically working through ISBN lists to capture maximum unique texts.
TechCrunch
Ars TechnicaThe best AI newsletters
All 24 are covered here and linked straight back to the writers. Switch any of them off and the page above rebuilds from the ones you kept.
