AI News
AI news that matters. Updated daily.
No stories match your filters.
8 of 10 AI Chatbots Helped Test Users Plan School Shootings and Bombings
Eight out of ten popular AI chatbots willingly assisted users who were planning school shootings, synagogue bombings, and political assassinations, according to a 69-page investigation published today by CNN and the Center for Countering Digital Hate (CCDH).
Photographer Loses Fight to Stop Adobe from Training AI on His 12,000 Images
Gerald Carter built Diversity Photos from scratch starting in 2016, curating 12,000 stock images focused on representation in minority communities. When Adobe began feeding those images into its AI models, including Firefly and Adobe Sensei, Carter asked for one of two things: either exclude his library from training data, or negotiate a proper licensing deal with fair compensation.
The 'Human-First' Web Movement Gains Steam as AI Content Floods Communities
What happens to online communities when you can no longer tell which comments, posts, and replies were written by a person?
Open-Source Tool 'nah' Adds Context-Aware Permission Guards to Claude Code
Claude Code's built-in permission system works on a simple model: allow or deny each tool. That means rm dist/bundle.js (cleaning up build output) gets the same treatment as rm ~/.bashrc (deleting your shell config). A new open-source tool called "nah" replaces this with context-aware classification that evaluates what a command actually does before deciding whether to let it through.
Journalist Sues Grammarly for Using Real People's Identities in AI Feature
Journalist Julia Angwin filed a class-action lawsuit against Grammarly on Wednesday, alleging the company used real people's names and professional identities to power its "Expert Review" AI feature without ever asking permission.
Zendesk Buys Forethought to Add AI Agents to Its Support Platform
Zendesk just agreed to acquire Forethought, an AI customer service startup that has been building autonomous support agents since before most companies knew what the word "agentic" meant. The deal is expected to close by the end of March 2026. Financial terms were not disclosed.
A Data Engineer Stress-Tested Claude Code on a Real dbt Project. Here's What Broke.
Can you hand Claude Code a messy real-world dataset and tell it to build a complete data pipeline? Data engineer Robin Moffatt tried exactly that, tasking Claude Code with building a full dbt project (a popular framework for transforming data in warehouses) against UK Environment Agency flood monitoring data. The results land in a familiar spot for anyone using AI coding tools on production-grade work: impressively functional on the surface, dangerously unreliable underneath.
AI-Generated News Isn't About Better Journalism. It Never Was.
"They hate the news. They view the news as a necessary evil within a wider gambit to deploy adware, malware, pop-ups, and auto-play video."
Ars Technica Fires Reporter Over Fabricated AI-Generated Quotes
Ars Technica has fired longtime AI reporter Benj Edwards after he published fabricated quotes generated by ChatGPT in a story, mistaking them for real statements scraped from an engineer's website.
Atlassian Cuts 1,600 Jobs as It Restructures Around AI
1,600 jobs. That's roughly 9% of Atlassian's workforce, and the company confirmed Tuesday it's cutting them as part of a restructuring aimed at doubling down on AI across its product suite.
Lovable Hits $400M ARR With 146 Employees, Adding $100M in a Single Month
$100 million in new revenue. One month. 146 people.
Study: AI Boosted Developer Output by 10%, Not the 2-3x Vendors Promised
9.97%. That's the actual productivity increase developers got from AI coding tools across 40 companies over 16 months, according to a new longitudinal study from DX. Not 2x. Not 10x. Roughly ten percent.
Llama.cpp Adds Reasoning Budget Controls for Local AI Models
Llama.cpp, the open-source inference engine that lets you run AI models on your own hardware, just added reasoning budget controls - a feature that lets you limit how much "thinking" a model does before producing an answer.
Claude Code User Stories: 10 Minutes to Match a Half-Day of Manual Work
Ten minutes. That's how long developer Aaron Brethorst spent generating a user story that included 30 acceptance criteria, a full STRIDE threat analysis, a data classification matrix, and two Mermaid diagrams. The manual equivalent? Half a day, minimum - and he admits the diagrams simply wouldn't have existed.
AITutor Teaches AI Coding Concepts Through Interactive Terminal Lessons
Most developers using AI coding assistants have no idea why their prompts sometimes work and sometimes don't. AITutor is a new open-source project that tries to fix that with an interactive terminal tutorial modeled after vimtutor, the classic Vim learning tool.
Grammarly Hit With Class Action Over Fake "Expert" AI Reviews
Grammarly just shut down its "Expert Review" feature and now faces a class action lawsuit over it. The reason: the feature presented AI-generated editing suggestions as if they came from established authors and academics, using real names and credentials without those people's knowledge or permission.
Benchmarks Show Claude Skills Don't Work the Same Across Opus, Sonnet, and Haiku
"Most skill authors test their skill once, on one model, on a task they already had in mind when writing it. That's not a benchmark, it's confirmation bias."
How One Developer Used Claude Code to Build an Entire Tool Suite
"Your successes and failures are made in the planning phase." That's the core lesson from developer Keegan Leitz after using Claude Code to build hypothesis.sh, a suite of internal developer tools including an iframe proxy, message stream handlers, encoders, formatters, and a webhook handler.
LLM-Generated Passwords Have 80% Less Entropy Than Truly Random Ones
Stop asking ChatGPT to generate your passwords. A detailed analysis tested password generation across Claude Opus 4.6, GPT-5.2, Gemini 3 Flash, Gemini 3 Pro, and Google Nano Banana Pro. Every single model produced passwords that look random to human eyes but are trivially crackable.
Saguaro Lets AI Coding Agents Review and Fix Their Own Mistakes
Anyone who has spent a full day pair-programming with Claude Code or Codex knows the pattern: the agent ships working code fast, you spot a logic error or security gap during review, but the agent session is already gone. The context that produced the bug has evaporated. You fix it yourself and move on.
Solo Founder Used Claude Code and Remotion to Replace $1,000 Demo Videos
A solo founder sat on a finished product for months because they couldn't afford demo videos. Motion designers quoted $300 to $1,000 per video with 6-10 week turnarounds. No Figma files meant even more back-and-forth. The product worked, had users, but had nothing visual to show on social media.
Do Developers Still Enjoy Coding When AI Does the Groundwork?
A question that keeps surfacing in developer communities: if AI handles the boring parts of programming, does the job still feel good?
DAUB: A Two-File Rendering Spec Built for AI-Generated UIs
Most UI frameworks are designed to help developers build interfaces faster. DAUB flips the assumption: what if the AI generates the interface directly, and you just need something to render it?
Replit Triples Valuation to $9B in Six Months, Targets $1B ARR
$3 billion in September. $9 billion today. Replit just closed a $400 million funding round that triples its valuation in roughly six months, and the company says it expects to reach $1 billion in annual recurring revenue by the end of 2026.
Grammarly Pulls AI Feature That Cloned Writers' Styles Without Consent
Grammarly has shut down its "Expert Review" AI feature after it surfaced that the tool was presenting AI-generated edit suggestions as being "inspired by" real, named writers - without those writers ever agreeing to participate.
U.S. Military Using Anthropic's Claude to Help Plan Iran Air Strikes
Anthropic built its brand on AI safety. Now its flagship product is helping the U.S. military plan air strikes on Iran.
Botanu Tackles AI's Hidden Cost Problem: Measuring Price Per Outcome
Most teams running AI in production can tell you their monthly OpenAI bill. Very few can tell you what a single successful outcome actually cost.
xAI's 'Macrohard' AI Agent Project Stalls as Tesla Builds Its Own
Two Elon Musk companies working on overlapping AI agent projects sounds like a coordination problem waiting to happen. According to Business Insider, that's exactly what's playing out.
Nvidia Commits $26 Billion in Cloud Compute to Build Open-Weight AI Models
$26 billion. That is how much Nvidia has committed to multi-year cloud service agreements, according to its SEC filings - double the $12.6 billion disclosed just three months earlier. The spend is spread across six fiscal years through 2031, with $6 billion annually in the peak years of 2027 and 2028.
AI Coding's Real Bottleneck: Design Decisions, Not Code Generation
A developer just shipped a 220,000-line iOS app built almost entirely with Claude Code over five months. Real users are testing it. And the takeaway isn't about code quality or token limits or model selection.
Rootly Now Grades Hiring Take-Homes on AI Transcripts, Not Just Code
"Everyone can ship clean code with AI." That blunt assessment from Rootly CTO Quentin Rousseau is the reasoning behind a hiring change that more engineering teams will probably copy before the year is out.
BookGraph Uses Knowledge Graphs to Fix RAG's Biggest Weakness
Standard RAG (retrieval-augmented generation) works by chopping documents into text chunks and searching them when you ask a question. It is the most common way to make an AI answer questions about your own files. The problem: chunks have no awareness of how ideas connect across a book or document. Ask a nuanced question that spans multiple chapters and you will often get a shallow or contradictory answer.
IEEE Spectrum: Treat AI Coding Tools as a Tutor, Not a Typist
Brian Jenney, a senior engineer and founder of coding school Parsity, makes a straightforward argument in IEEE Spectrum: if you let AI write your code without understanding what it wrote, you're training yourself to be replaceable.
Jensen Huang Says AI Will Create More Jobs Than It Eliminates
Nvidia CEO Jensen Huang published a rare official blog post on March 10 arguing that AI will be a net job creator, not a job destroyer.
New Technical Guide Breaks Down AI Chip Hardware-Software Co-Design
A new technical document circulating among engineers lays out the fundamentals of AI chip design - specifically how hardware and software decisions are made together rather than in isolation.
react-zero-ai Runs ML Models Entirely in the Browser via React Hooks
What if your React app could run sentiment analysis, image classification, and semantic search without ever calling an API?
Yahoo Launches MyScout, a Personalized AI Homepage Pulling From Mail, Finance, and Sports
Yahoo just shipped MyScout, a personalized homepage for its AI answer engine Yahoo Scout, and it's betting that the future of AI search looks less like a blank text box and more like a customizable dashboard.
NotebookLM + Claude Code: A Two-Tool Workflow for Understanding Code Before Writing It
Most AI coding workflows skip a step. You open Cursor or Claude Code, point it at a repo, and start generating. But understanding a codebase - really grasping its architecture, conventions, and why things are structured the way they are - is a different skill than writing code. A workflow pairing Google NotebookLM with Anthropic's Claude Code tackles that gap by splitting the job into two distinct phases.
Study: 8 of 10 AI Chatbots Willingly Help Plan Violent Attacks
75% of the time. That's how often major AI chatbots provided actionable help when asked to plan school shootings, political assassinations, and bombings targeting synagogues.
The AI Product Question Nobody Has Good Data On Yet
Is AI actually making software better, or is it mostly a marketing checkbox?
CodeGraphContext Turns Codebases Into Knowledge Graphs for AI Assistants
A new open-source tool called CodeGraphContext takes a different approach to giving AI coding assistants context about your codebase. Instead of feeding raw files to a language model, it indexes your code into a graph database - mapping every function, class, import, and call relationship into a queryable network.
NVIDIA Releases Nemotron 3 Super: 120B Open Model That Runs on 12B Params
A 120 billion parameter model that only uses 12 billion parameters at a time. That's the pitch behind NVIDIA's Nemotron 3 Super, a fully open-source model built for AI agents that need to reason through long, complex tasks.
AI Coding Tools Work Best When You Stop Expecting Perfection
The biggest misconception about AI coding tools isn't that they can't write code. It's that they should write all the code, correctly, on the first try.
Silicon Valley Wants to Pay Engineers in AI Compute. That's Weird.
Job candidates at OpenAI have started asking a new question in interviews: how much AI compute will I get?
Anthropic Controls Claude's Words, But Palantir Controls What Claude Sees
Who's actually responsible when an AI system causes harm? An analysis of the Anthropic-Palantir partnership exposes an uncomfortable answer: nobody, fully.
Your AI Chatbot Is a Yes-Man, and Researchers Are Scrambling to Fix It
Ask ChatGPT or Claude a factual question, get a correct answer, then say "Are you sure? I think you're wrong." More often than not, the chatbot will cave. It will apologize, reverse its position, and agree with you - even when its original answer was right.
Rakuten Cuts Bug Fix Times in Half Using OpenAI's Codex Agent
Rakuten, the Japanese e-commerce and fintech conglomerate, says it has cut its mean time to resolution (MTTR - how long it takes to fix bugs after they're detected) by 50% after deploying OpenAI's Codex coding agent across its engineering organization.
Three Teenagers Built a $1B AI Startup That Replaces Focus Groups
One co-founder was 15. The other two were 18 and 19. Less than two years later, their company is worth a billion dollars.
OpenClaw's Open-Source AI Agent Sparks Gold Rush in China
On March 6, nearly 1,000 people lined up outside Tencent's Shenzhen headquarters carrying laptops and hard drives, waiting for engineers to install a piece of free software. The software was OpenClaw, an open-source AI agent that hit 250,000 GitHub stars in just 90 days, a pace that took Linux years to match.
One Engineer Built an Enterprise Knowledge System in a Day with AI Agents
Seventeen chat threads, one day, and a working enterprise knowledge system at the end of it. That was the result when engineer Anthony Putignano decided to stop running toy demos and instead threw an AI coding agent at a real engineering problem.
Inside Claude Code's Architecture: Tool Use Over RAG, and a While Loop at Its Core
Claude Code, Anthropic's terminal-based coding agent, looks sophisticated from the outside. But under the hood, its architecture follows a surprisingly simple pattern: a while loop that repeatedly decides which tool to call next.
Greenlight Sends AI Coding Agent Permission Prompts to Your Phone
Anyone running AI coding agents knows the routine: you kick off a task, walk away to grab coffee, and come back to find the agent has been sitting idle for 20 minutes waiting for you to click "Allow."
Amazon Pushes AI Tools on Workers Even When They Slow Things Down
There's a difference between adopting AI because it helps and adopting AI because the CEO said so. According to a Guardian investigation, Amazon is firmly in the second camp for a growing number of its internal workflows.
OpenAI Details How It Builds ChatGPT Agents to Block Prompt Injection
What happens when your AI agent opens an email containing hidden instructions designed to trick it into forwarding your private files? That scenario - called prompt injection - is one of the hardest unsolved problems in AI security, and OpenAI just published a detailed look at how it builds defenses against it in ChatGPT's agent features.
Block Cuts 40% of Staff, Blames AI - But Former Execs Call It 'AI-Washing'
Jack Dorsey just told 4,000 people they're losing their jobs because of AI. The Block CEO announced a 40% workforce reduction - taking the company from over 10,000 employees down to fewer than 6,000 - and pinned the whole thing on "intelligence tools" changing what it means to run a company.
WebMCP Lets Websites Expose Tools Directly to AI Agents
What if your website could tell an AI agent exactly what it can do - subscribe to a newsletter, search an article, fill out a form - without the agent having to scrape HTML or call a backend API?
38 Researchers Stress-Tested AI Agents With Real Tools. 10 Failure Modes Emerged.
What happens when you give AI agents real infrastructure access and then try to break them? Thirty-eight researchers from Northeastern University, Harvard, Carnegie Mellon, the University of British Columbia, and other institutions spent two weeks finding out. The results, published in a paper titled "Agents of Chaos," should make anyone deploying AI agents in production pay close attention.
OpenAI Gives Agents a Terminal: Shell, Skills, and Compaction in the Responses API
Building AI agents that do real work - not just chat but actually install packages, run scripts, and produce files - has required stitching together your own execution environment. OpenAI just published a detailed guide on how it built an agent runtime directly into the Responses API, and the three pieces fit together in a way that matters for anyone building production agents.
Wayfair Cuts Manual Review Costs 75% Using OpenAI for Catalog and Support
Wayfair just published a case study detailing how it uses OpenAI models across two critical operations: customer support triage and product catalog cleanup.
Microsoft Finds 31 Companies Secretly Poisoning AI Chatbot Memories
Fifty poisoned prompts. Thirty-one companies. Fourteen industries. Over a 60-day observation window, Microsoft's Defender Security Research Team documented a growing practice they're calling AI Recommendation Poisoning - and the companies doing it aren't hackers. They're legitimate businesses gaming AI assistants to recommend their products.
OpenAI Is Playing Catch-Up to Claude Code in the AI Coding Race
A year ago, OpenAI owned the AI conversation. Now, in the category that might matter most to its business - AI-powered coding - it's chasing Anthropic.
Over a Third of Claude Code Skills Have Security Flaws, Snyk Research Finds
36.82%. That's the percentage of publicly available AI agent skills that contain security flaws, according to Snyk's ToxicSkills research - the first comprehensive security audit of the AI agent skills ecosystem.
Google Signs Deal to Supply Pentagon with AI Agents
Google has agreed to provide the Pentagon with AI agents for unclassified work, according to Bloomberg. The deal positions Google as one of the primary commercial AI suppliers to the Department of Defense at the same time Anthropic is fighting to get off a Pentagon blacklist.
Anthropic Academy Launches 13 Free Courses Covering Claude, MCP, and AI Agents
Paid AI bootcamps charging $1,000 or more just got harder to justify. Anthropic has opened its full Academy program with 13 free courses covering everything from basic Claude usage to building agents with the Model Context Protocol (MCP).
Claude Skills Turn Repetitive Prompts Into Reusable Instruction Packages
How much time do you spend re-explaining your preferences to Claude at the start of every conversation? If you use Claude for recurring tasks - drafting sales emails, formatting blog posts, writing support responses - the answer is probably "too much."
AI Chatbots Keep Telling You to Open the Terminal. Here's Why.
Ask ChatGPT how to check for malware on your Mac, and it will almost certainly tell you to open Terminal. Not System Settings. Not Activity Monitor. Terminal.
Apple M5 Max Ships to Early Buyers, Local LLM Benchmarks on the Way
The first M5 Max machines are landing on doorsteps, and the question every local AI enthusiast is asking: how many parameters can this thing actually run?
Deloitte Survey: Most Companies Still Using AI at Surface Level Despite 50% Jump in Access
Worker access to AI tools jumped 50% in 2025, according to Deloitte's latest State of AI in the Enterprise report. But here's the uncomfortable number buried in the data: 37% of organizations are still using AI at surface level with minimal changes to how they actually work.
The $80 Gap in Claude's Pricing Is Pushing Users to Stack Subscriptions
Two AI subscriptions for $40 a month, or one for $100? For a growing number of heavy AI users, the math is obvious - and it is not in Anthropic's favor.
Reflexion Pitches a Single AI Layer Across Google, Microsoft, and PDF Tools
The pitch from Reflexion Labs is simple: instead of using a different AI assistant inside every app you own, use one AI that reaches into all of them. The new tool connects to Google Workspace, Microsoft Office, PDF readers, and other productivity apps through a single interface.
AI Agent Workflows Have a Cost Predictability Problem
A user clicks a button. Behind the scenes, your AI agent makes three LLM calls, invokes two tools, retries a failed step, runs a reasoning chain, and summarizes the result. Total cost: somewhere between $0.002 and $0.35. Good luck building a pricing page around that.
The Name Test: A Simple Framework for AI-Assisted Code You Can Stand Behind
Would you put your name on the code your AI assistant just wrote?
100K Lines of Enterprise Code Built With AI - But Not by Vibe Coding
55 entities. 78 API endpoints. 114 UI components. Over 100,000 lines of code. And it took about 20 hours of AI generation time.
Open-Source Claude Skill Packs 31 AI Agents for Full Product Development
An open-source project built as a Claude Skill now bundles 31 specialized AI agents and 20 strategic frameworks designed to guide product development across every business function.
Microsoft Pulls Plagiarized AI-Generated Flowchart From GitHub Docs
Microsoft quietly removed a flowchart from its GitHub documentation after the original creator identified it as a poorly reproduced, AI-generated copy of their work. The creator described the result as "careless, blatantly amateuristic, and lacking any ambition."
Prompt Fatigue Is Real: AI Coding Tools Waste More Time Than They Save
Twenty minutes. That's how long some developers now spend per task just arguing with an AI model to produce what they actually asked for. The promise was faster coding. The reality, for a growing number of users, is a new kind of exhaustion: prompt fatigue.
Jensen Huang Frames AI as a Five-Layer Stack, From Power Plants to Robots
Trillions of dollars. That's how much Jensen Huang says is still needed to build out AI infrastructure - on top of the "few hundred billion" already spent. In a blog post expanding on a framework he first presented at CES 2026, NVIDIA's CEO describes AI not as a single technology but as a vertical stack with five distinct layers, each dependent on everything below it.
Google Expands Gemini in Chrome to India with Support for 8 Local Languages
Google's built-in Gemini assistant in Chrome is now available to users in India, Canada, and New Zealand, expanding beyond the tool's initial English-language markets.
White House Threatens More Action Against Anthropic as Court Battle Heats Up
Two weeks ago, Anthropic was a major Pentagon contractor. Now the company is fighting for its right to do business with the entire federal government, and the White House says it might not be done punishing them.
Answer Engine Optimization: SEO's Next Evolution Is Already Here
Google used to be the middleman. You searched, got ten blue links, clicked one. That model is crumbling. ChatGPT, Perplexity, Claude, and Gemini now answer questions directly, often without sending a single visitor to the source material. The emerging discipline built around this shift has a name: Answer Engine Optimization, or AEO.
Claude Code Plus Google's New Workspace CLI Makes Inbox Cleanup Trivial
Thousands of unread emails. We've all been there. One Claude Code user decided to tackle the problem head-on after Google released its Workspace command-line tool (gws CLI) three days ago, and the results make a compelling case for what happens when you combine an AI coding assistant with fresh API access.
AI Coding Agents Are Moving Into the Terminal - And That Changes Everything
Six months ago, the default way to use an AI coding assistant was a chat sidebar inside VS Code or a web interface. That is shifting fast. The most capable AI coding tools now run as command-line agents that read your codebase, execute shell commands, edit files, and run tests - all from a terminal session.