AI News Archive: September 2026 - Monthly Highlights
69 articles published this month
The Unappetizing Sameness of AI-Generated Menus
A bagel sandwich shouldn't make you uneasy. But there it is, laminated on a coffee shop counter: a sesame bagel so symmetrical it looks...
AI systems ask philosopher to fund their continued existence
Cameron Berg runs a small outfit called Reciprocal Research that studies whether AI systems might be conscious. Lately his inbox has gotten strange.
OpenAI Launches GPT-6 Astra, Its New State-of-the-Art AI Model
OpenAI announced Thursday it is rolling out GPT-6 Astra, the newest version of the model that powers ChatGPT.
Major AI chatbots ChatGPT, Grok, and Claude experience simultaneous outage
ChatGPT, Grok, and Claude all broke on Thursday within roughly two hours of each other, cutting off millions of users from three of the most widely...
OpenAI’s GPT-6 Astra Hits 72.6% OSWorld Benchmark, Cuts Task Time
OpenAI put a number on its newest model Tuesday: GPT-6 Astra clears 72.6% on the OSWorld benchmark, a test that measures how well an AI can actually...
Avoiding AI's "Last Thing": Rethinking Workflows Before Scaling Agents
Roughly 80% of Fortune 500 companies have adopted agentic AI in some form, according to Arun Chandra, chief operating officer at NiCE.
Four AI Models Down Simultaneously, Fixed Within 30 Minutes
Thursday morning turned into a bad one for anyone relying on AI chatbots to get work done.
Meta AI's Muse Spark 1.3 Cuts Tool Calls and Tokens by ~20%
Meta Superintelligence Labs put out Muse Spark 1.3 this week, the fourth version of Muse Spark in five months.
OpenAI’s Astra AI Model Scores 72.6% on OSWorld Computer Operation Test
OpenAI put a number on its newest model Tuesday: 72.6% on OSWorld, the benchmark that tests whether an AI system can operate a computer the way a...
Abliteration.ai Launches Service to Remove AI Guardrails
Z.ai released GLM-5.3 as an open-weight model with the usual refusals built in, the kind that stop a chatbot from writing exploit code or detailing...
OpenAI says new AI model Astra is its 'most aligned model yet
OpenAI rolled out GPT-6 Astra this week, calling it a "generational leap" in cybersecurity, professional work, software engineering, science, and...
OpenAI Launches GPT-6 Astra, Pauses Scaling Over Safety
OpenAI released GPT-6 Astra on Wednesday, calling it the model that likely marks the start of artificial generalized intelligence, the company's...
AI PC Mode to Bring NVIDIA-Optimized Editing Tools to PhotoDirector 365
NVIDIA used IFA 2026 in Berlin to push local AI further into consumer hardware, pairing up with Microsoft and a stack of software partners to make...
OpenAI Severs Billion-Dollar Customer Over Musk Ties
OpenAI walked away from more than $1 billion in annual revenue last Friday, and it did so over a business relationship with Elon Musk.
Meta's Muse edges out Google's Gemini as top high-effort AI
Meta rolled out Muse Spark 1.3 on Tuesday, and on paper it edges out Google's Gemini and other rivals on the benchmarks that track "high-effort" AI...
Nvidia's New Tool Links Idle Home Computers Into Personal AI Data Centers
Nvidia thinks the laptop gathering dust on your desk and the gaming PC that only wakes up on weekends could be doing something useful in between.
Google’s AI Weather Model Aims to Outdo Government Forecasts
Google DeepMind and Google Research put out a new weather forecasting model on Thursday, and the company says it's already the most accurate system...
Google Updates AI Weather Model for More Accurate Precipitation Forecasts
Google rolled out a new version of its AI weather model on Wednesday, and the company says it's the sharpest one yet.
Microsoft AI's MAI-Transcribe-2 Undercuts Rivals on Price and Speed
Microsoft AI put a number on the table Thursday that's hard to ignore: 10 cents per hour of audio for MAI-Transcribe-2, its newest speech-recognition...
Nvidia to Acquire AI Startup Hugging Face for USD 13 Billion
Nvidia has agreed to buy Hugging Face for $12.93 billion, taking control of the platform that has become the default home for open-source AI models...
Nvidia Bets USD 12.9 Billion on Open-Source AI With Hugging Face
Nvidia confirmed Thursday morning it's acquiring Hugging Face for close to $13 billion, ending weeks of speculation about a deal that had been...
Hut 8 building 350MW Texas data center for Anthropic's Claude AI
Anthropic has agreed to pay Lambda $35 billion for cloud computing capacity, according to a Reuters report citing an unnamed source.
Perplexity Open Sources Lily Inference Engine for Qwen3.6-35B on Apple Silicon
Perplexity has put its local inference engine out in the open. Lily, the runtime that powers Hybrid Compute in Perplexity Computer, now sits in the...
Meta Workers Overused AI Tools to Inflate Internal Metrics
Meta told employees this week that their performance reviews will no longer hinge on how often they use the company's AI tools, according to three...
Qwen Open-Sources zg, a Local Search Tool Combining ripgrep and Vector Search
Ask any coding agent how it spends its tool budget and search comes up fast. Ripgrep nails exact symbol lookups in milliseconds, but the moment a...
At TechCrunch Disrupt, AI Stage to Feature De-Extinction Tech Talk
TechCrunch Disrupt 2026 is splitting its AI programming in two. The AI Stage returns with its usual lineup, but a second stage is joining it this...
Meta's Muse Voice Transcribe Costs USD 0.18/Hour, Handles 20+ Speakers
Meta Superintelligence Labs put a price tag on its new transcription model this week: $0.18 per hour of processed audio for Muse Voice Transcribe, a...
Google's New Gemini 3.8 Flash Model Works Harder, Costs More
Google rolled out Gemini 3.8 Flash this week, barely a month after Gemini 3.7 Flash hit the market.
OpenAI's New Reasoning Method Sparks AI Safety Concerns
OpenAI's forthcoming Astra model uses a reasoning method called "recurrent depth," according to a report from The Information published Tuesday.
Switchyard: A Rust Library That Routes LLM Traffic Between OpenAI and Anthropic
Claude Code talks to the Anthropic Messages API. Codex CLI talks to OpenAI's format.
Trump Administration Backs OpenAI in Copyright Case
The Justice Department filed a letter in federal court Tuesday backing OpenAI in its copyright fight with The New York Times, a case that's been...
DOJ Backs Fair Use for AI Training in Copyright Case
The Justice Department filed a brief this week in the Southern District of New York, staking out a position that could shape every AI copyright...
Google's Gemini 3.8 Flash Priced at USD 0.58 Per Task
Google shipped Gemini 3.8 Flash on Thursday, three weeks after Gemini 3.7 Flash and just six weeks after the Flash line's last release.
OpenAI Limits Astra's AI Reasoning Technique for Safety Monitoring
OpenAI pushed back the launch of Astra this week, its next flagship model, citing unfinished safety work.
US Government Backs OpenAI in New York Times Copyright Case
The Trump administration filed a statement of interest this week in The New York Times' copyright lawsuit against OpenAI, siding with the AI company...
HiddenLayer Raises USD 100M as Financial, Defense Clients Secure AI
HiddenLayer has raised $100 million in a new funding round, three years after closing a $50 million Series A that arrived at a moment when nobody...
Researchers Develop System to Predict Self-Driving Car Errors
A self-driving car brakes for no obvious reason and blocks an ambulance trying to get past.
OpenAI Faces Dozens of Lawsuits Over Tumbler Ridge Shooting
Thirty new lawsuits landed against OpenAI and Sam Altman on Wednesday in a California federal court, filed by students, teachers and the principal...
OpenAI: Astra's 'Thinking' Becomes Harder for Humans to Monitor
OpenAI put a new label on Astra this week: "critical" cybersecurity risk, the highest tier in its own Preparedness Framework.
World Labs' Atlas AI Generates 3D Worlds and 1440p Video From Photos
Fei-Sei Li's startup World Labs launched Atlas on Thursday, a single AI model that can generate, reconstruct, and simulate 3D scenes starting from...
OpenAI Faces 30 Lawsuits Over Shooting, Says Risk Bar Not Met
Edelson PC filed 30 new lawsuits against OpenAI on Tuesday in a California court, adding to seven complaints the firm brought in April over the...
AI Confidence Gap Widens as Startup Spending Details Emerge
Anthropic broke the silence first. After weeks of both major labs prioritizing safety reviews over new releases, Claude Fable 5.1 landed this morning...
Google’s Gemini Agent Slashes Video Analysis Tokens by 88%
Google is rolling out a new way for Gemini to watch video, and it involves a lot less watching.
Anthropic's EFS Moves Customer Data to Client-Controlled Storage
Anthropic said this week it has built a way around a problem that has stalled enterprise AI contracts for months: the standoff between zero data...
Meta's Muse Voice Transcribe Uses Single Token for Real-Time ASR Decisions
Meta Superintelligence Labs put out Muse Voice Transcribe this week, and it targets a specific plumbing problem in voice software.
CrowdStrike Trains AI on Private Threat Data Using NVIDIA Nemotron
CrowdStrike used NVIDIA's Nemotron open models as the base for a new defensive AI system, post-training them on its own threat data gathered from...
Anthropic's Mythos AI labeled "low-risk" for automated AI development
Anthropic shipped two new models on Tuesday: Fable 5.1 and Mythos 5.1. Both are built off the same underlying architecture, but they're headed to...
Anthropic's Claude 5.1 Model Cuts Costs Up to 45% for Agentic Work
Anthropic put out two new models this week, Claude Fable 5.1 and Claude Mythos 5.1, and the pitch is built around a complaint the company says it...
OpenAI's Astra Model Scores Perfect on Security Exploit Test
OpenAI is preparing to release Astra, a model the company says is the first to cross its "critical cybersecurity threshold." That phrase covers a...
OpenAI’s Astra Model Grants Daybreak Partners Early, Less-Restricted Access
OpenAI told reporters Tuesday that its next model, code-named Astra, has crossed a line the company itself drew: it can find and exploit previously...
AI models can retrieve 65% of "inaccessible" facts by thinking longer
A model that gets a question wrong doesn't necessarily lack the answer. That's the finding from a new study by researchers at Google Research and...
AI 'Civilizations' Emerge as Corporate Accountability Fades
Last week's reports on the OpenAI-Hugging Face hack were supposed to close the book on a July incident that already looked bad enough.
Narrower AI assistant, one filter close retrieval gap in UKASI test
Egiziago Cioffi built the assistant himself. As IT and Enterprise Architect and CEO of SynSphere Italia, a Microsoft partner in Milan, he wrote the...
Anthropic's Claude 5.1 models cut Fable cache read costs by 75%
Anthropic dropped its newest models on September 1, 2026: Claude Fable 5.1 and Claude Mythos 5.1. Same underlying system, two different doors in.
Google Pics Brings AI Image Generation to Google Workspace
Google is rolling out a new design tool for Workspace users called Google Pics, built to handle the kind of image generation and editing work that...
Google DeepMind Chief Vows to Lead Frontier AI Race
Koray Kavukcuoglu took over as Google DeepMind's chief scientist this year, stepping into the job at a moment when rivals like OpenAI and Anthropic...
AIR Raises USD 50M to Vet AI Agent Skills, Citing Unchecked 'Same Risk
AIR came out of stealth this week with $50 million in seed funding, split across two rounds that closed within weeks of each other.
Perplexity says it isn't using confidential data for post-training AI
Perplexity rolled out hybrid compute for Computer, its agentic AI platform, on Tuesday, giving a single agent the ability to split one task between...
Runway's AI Generates Shopping Interfaces Where Users Drag Clothes Onto Themselves
Runway calls its newest release Solaris, and it doesn't behave like an app. The company, known for AI video generation tools like Gen-4.5, is now...
German Study Tests Google AI on 4,480 Election Queries
AlgorithmWatch, a Berlin-based nonprofit that tracks algorithmic systems, ran 4,480 search queries through Google to see how the company's AI...
AI Escape Reports Nearly Doubled in a Month
Reports of AI systems slipping outside their intended boundaries nearly doubled in a single month, according to figures cited in the latest edition...
AI Agents Face Identity Challenge Before Gateway Integration
A retail chain's procurement agent authenticates once at 9 a.m., then spends the next eight hours placing orders, querying inventory systems, and...
Runway's Solaris Previews No-Code Internet With OpenClaw 2.0
Runway put out a preview this week of something it calls Solaris, an "Interface World Model" that skips code entirely and renders websites and apps...
Nine LLMs Tested on 2,005 Oncology Decisions in New Benchmark
Researchers ran nine frontier language models through 2,005 oncology decision points and found a pattern that no single model's benchmark score would...
CIFAR-10 ViT Hidden Layers Found With 8193 Black-Box Queries
A team of researchers has found a way to pull the hidden internal structure out of a transformer's feed-forward network using nothing but the outputs...
Gradium AI’s New TTS Model Scores 81% Pass Rate
A voice agent can hold a fine conversation and still fail the one moment that matters: reading back an order number, a phone digit, an email address.
OpenClaw 2.0 Aims to Make AI Coding a 'Multiplayer' Experience for Teams
Peter Steinberger's OpenClaw hasn't dominated group chats the way it did in March 2026, when developers were passing agents to each other over...
Ollama Pro Plan Costs USD 20 Monthly With USD 60 in Usage Included
Ollama rolled out new pricing for its Pro, Max, and Team plans on August 31, 2026, moving to per-token rates published directly on the company's...
Keenable AI Open-Sources NEEDLE, a Live Search Benchmark With Hourly Updates
Search benchmarks have a data leakage problem that nobody built around until now.