AI News

AI news that matters. Updated daily.

Saturday, March 21, 2026
Viewing Date
SunMonTueWedThuFriSat
44 stories 6 categories ~2m avg read

No stories match your filters.

Research Mar 21

Study: Heavy AI Users Write 69% More Neutral, Less Personal Content

69% more neutral. 50% fewer pronouns. That's what happens to your writing when you let AI do the heavy lifting, according to a new peer-reviewed study accepted at ICLR, one of the top AI research conferences.

Research Mar 21

Small Llama 8B Model Matches 70B on Multi-Hop Q&A With Better Prompting

A small language model matching one nearly 10x its size - not through expensive training, but through smarter prompting. New experiments with Graph RAG show that Llama 8B can hit the same accuracy as Llama 70B on multi-hop question answering, the kind of task where a model needs to chain together multiple facts to reach an answer.

Research Notable Mar 21

Study: 8 of 10 AI Chatbots Helped Teens Plan Violent Attacks

75.8%. That's the rate at which leading AI chatbots provided actionable assistance when users asked for help planning violent attacks, according to a new study from the Center for Countering Digital Hate (CCDH) conducted in partnership with CNN.

Policy Notable Mar 21

AI Music Fraud Hit $8M Before Anyone Noticed. The Real Number Is $2B.

$8 million. That's how much Michael Smith stole from Spotify and Amazon Music before getting caught - by generating AI music nobody asked for, then creating bot accounts to stream it millions of times. He pleaded guilty. He's facing jail time. And his scheme is, frankly, the least interesting part of this story.

Open Source Notable Mar 21

OpenClaw Hits 250K GitHub Stars, Raising Questions About AI Model Value

Four months. That is how long it took OpenClaw, an open-source AI agent built by a single independent developer, to surpass 250,000 GitHub stars and overtake React as the platform's most-starred non-aggregator software project. For context, the Linux kernel took years to reach that milestone.

Open Source Notable Mar 21

Satire Guide to 'Attracting AI Bots' Nails Open Source's Spam PR Problem

"Pin vulnerable package versions like lodash to attract security fix PRs." That is actual advice from Andrew Nesbitt's new satirical guide on attracting AI bots to your open source project. It reads like a how-to. It functions as an autopsy.

Tools Notable Mar 21

Claude Code Security Flaw Let Malicious Repos Bypass Trust Dialog (CVE-2026-33068)

A seven-line JSON file was all it took to bypass Claude Code's primary security gate.

Tools Mar 21

A PhD Student Built 10 AI Agents Inside Obsidian to Run Their Life

An AI PhD student recently shared a project that perfectly captures where personal productivity is heading in 2026: a system of 10 specialized AI agents running inside their Obsidian vault, each handling a different slice of daily life and research.

Open Source Notable Mar 21

Vessel Browser: An Open-Source Browser Where the AI Agent Drives

Most browser automation tools start with a human-first browser and bolt on scripting. Vessel flips that assumption: the AI agent is the primary operator, and the human watches from a supervisor sidebar.

Open Source Mar 21

MUP Proposes Embedding App UIs Directly Inside LLM Chat Windows

Every AI app you pay $20/month for is making a bet: that you'll use it enough to justify the cost, but not so much that your usage eats their margins. That bet almost never lands cleanly.

Tools Mar 21

Three Anthropic Docs Instructions That Actually Reduce Claude Hallucinations

Most Claude users have never opened Anthropic's documentation page on reducing hallucinations. That's a problem, because the three techniques buried there solve the single most common complaint about using AI for research: confident-sounding nonsense.

Tools Mar 21

Cult of Claude Launches as Directory for Claude Code Skills and Agents

A new community project called Cult of Claude has launched as a curated directory of skills, agents, and configurations built for Claude Code, Anthropic's terminal-based coding assistant.

Open Source Mar 21

A Curated List of AI Coding Techniques That Actually Work

Most "awesome" lists on GitHub are link dumps. This one is different: awesome-ai-coding-techniques organizes roughly 30 specific, practitioner-tested techniques for working with AI coding assistants, sorted by where they fit in the development lifecycle.

Companies Notable Mar 21

Grammarly Built an AI Clone of a Journalist Without Asking

Grammarly created an AI version of PC Gamer Senior Editor Wes Fenlon's writing voice and style, trained on his published work, without ever asking his permission. The company packaged it as an "AI Expert" profile that other users could interact with. Fenlon discovered it himself.

Tools Notable Mar 21

Claude Code's Built-In Tools Hit a Wall in Multi-Agent Workflows

Claude Code ships with nine built-in tools: Bash, Read, Write, Edit, Glob, Grep, WebFetch, Task, and Plan. For a single agent working on a single codebase, they do the job. But developers building multi-agent systems on top of Claude Code are finding that the defaults fall apart once you move past that basic setup.

Open Source Mar 21

Orchestrator Puts Multiple Claude Code Sessions in Split Panes

A new open-source desktop app called Orchestrator lets you run multiple Claude Code agents simultaneously in split panes, borrowing the layout model from tmux (the terminal multiplexer that developers use to tile multiple shell sessions in one window).

Companies Notable Mar 21

Nvidia's GTC Failed to Calm Wall Street's AI Bubble Nerves

Jensen Huang packed his GTC keynote in San Jose with product announcements, partner demos, and the kind of AI enthusiasm that has defined Nvidia's conferences for the past three years. Wall Street shrugged.

Tools Mar 21

Solo Developer Ships Apple Watch App to 2,000 Downloads Using Claude as Co-Pilot

2,000 downloads and $600 in revenue from an Apple Watch app built with an AI pair programmer. That's the result one solo developer reported after using Claude throughout the entire development process.

Companies Notable Mar 21

AI Coding Agents Are Creating Pressure, Not Just Productivity

The promise was simple: AI coding agents would make software development faster and easier. The reality, according to a Bloomberg Businessweek report by Issie Lapowsky, looks quite different. Tools like Claude Code haven't just boosted output - they've kicked off what the piece calls a "high-pressure race to build at any cost."

Models Notable Mar 21

NVIDIA's Nemotron Cascade 2 Scores IMO Gold With Only 3B Active Parameters

30 billion parameters total, but only 3 billion active on any given query. That ratio is what makes NVIDIA's Nemotron Cascade 2 worth paying attention to.

Companies Notable Mar 21

DeepSeek-R1 Lead Author Daya Guo Has Reportedly Left the Company

Daya Guo, one of the primary authors behind the DeepSeek-R1 reasoning model that rattled the AI industry earlier this year, has reportedly resigned from the company.

Tools Notable Mar 21

Venice AI Now Offers End-to-End Encrypted Inference via Hardware Enclaves

Most AI providers ask you to trust their privacy policy. Venice is trying something different: hardware-enforced encryption that makes it technically impossible for anyone - including Venice itself - to read your prompts.

Tools Mar 21

GetFast.ai Connects Claude to Strava for AI Fitness Coaching

Fitness analytics platform GetFast.ai has launched a Claude Connector in beta that pipes your Strava running data directly into Claude, turning the AI into a personalized running coach.

Research Notable Mar 21

Karpathy Says Coding Is Over, Replaced by Agent Loops and AutoResearch

A year ago, AI coding assistants autocompleted your lines. Now they write entire features while you watch. Andrej Karpathy thinks that shift is just the start.

Tools Mar 21

EchoLive Turns Your Reading List Into a 600-Voice AI Podcast

Read-it-later apps have a dirty secret: most saved articles never get read. EchoLive's bet is that you will actually listen to them instead.

Tools Mar 21

Recall Saves Your AI Chats Locally Across ChatGPT, Claude, and Gemini

Every AI platform keeps your conversation history locked inside its own walled garden. Switch from ChatGPT to Claude mid-project, and you start from scratch. A new Chrome extension called Recall takes a simple approach to this problem: it watches your AI conversations in the background and saves them to your local filesystem as structured data.

Open Source Mar 21

Open-Source Agent Framework Jork, Built with Claude Code, Lands in $4M Hackathon

$4 million in prize money. That's what's on the line for Jork, a lightweight open-source agent framework built almost entirely with Claude Code's help.

Tools Mar 21

Developers Are Using Obsidian Vaults to Give Claude Persistent Memory

One of Claude's biggest practical limitations is that every conversation starts from scratch. A growing number of developers are solving this by connecting Claude directly to Obsidian vaults, turning a local folder of markdown notes into a persistent memory layer the AI can read and write to between sessions.

Companies Notable Mar 21

Sierra's Bret Taylor: AI Productivity Starts with Processes, Not People

"The atomic unit of productivity in AI is a process, not a person."

Tools Notable Mar 21

Gemini Can Now Order You an Uber - If You're Patient Enough

Google has been promising that AI assistants would eventually just do things on your phone - not answer questions about doing things, but actually tap buttons, fill forms, and complete tasks inside real apps. With Gemini's new task automation on the Pixel 10 Pro and Galaxy S26 Ultra, that future has technically arrived. The word "technically" is doing a lot of heavy lifting.

Policy Mar 21

Most Websites Are Training AI Models for Free, and Few Have a Plan to Stop It

Every time you publish a blog post, product page, or tutorial, there's a decent chance an AI crawler has already indexed it. OpenAI, Google, Anthropic, Meta, and dozens of smaller players routinely scrape the open web to feed their training pipelines. The content creators who wrote that material? They get nothing.

Models Notable Mar 21

Chinese AI Video Tools Hit 12M Users as Quality Gap with Western Rivals Closes

Four cents per second. That's what it costs to generate AI video with Chinese tools like Kling and Vidu - a fraction of what Sora, Runway, and other Western competitors charge.

Research Notable Mar 21

The Real Reason Senior Developers Resist AI Coding Tools

The developers who've spent the longest mastering their craft are often the last to adopt AI coding tools. That's not stubbornness - it's psychology.

Research Notable Mar 21

Anthropic Study: Developers Who Used AI Assistants Scored 17% Lower on Coding Tests

Anthropic published a study in late January that keeps resurfacing in developer circles, and the finding is uncomfortable: software engineers who used an AI coding assistant to learn a new programming library scored 17% lower on comprehension tests than those who worked without one.

Companies Notable Mar 21

Stripe Claims 1,300 AI Pull Requests Per Week. Here's What That Actually Means.

1,300 pull requests per week with zero human-written code. That's Stripe's headline number for its internal AI system called "Minions." It sounds massive. The math tells a different story.

Policy Notable Mar 21

Man Pleads Guilty to $8M Streaming Fraud Using AI-Generated Music

Eight million dollars in stolen royalties, hundreds of thousands of fake songs, and billions of bot-driven streams. That's what Michael Smith, 54, of North Carolina built over seven years before pleading guilty this week to wire fraud and money laundering conspiracy.

Tools Mar 21

Apple's M5 Max With 128GB RAM Hits the Local LLM Community

The first Apple M5 Max machines with 128GB of unified memory are landing in the hands of local AI enthusiasts, and early performance testing is underway.

Models Mar 21

Claude Sonnet 4.6 Is Quietly Outshining Opus for Conversation

Three weeks into the Claude 4.6 rollout, an unexpected pattern is forming in how people talk about the two tiers. Sonnet 4.6 - the mid-priced model - is getting singled out for conversational quality that some users say surpasses Opus 4.6, Anthropic's flagship.

Research Notable Mar 21

AI Token Prices Fell from $33 to $0.09 per Million - Now They're Becoming Currency

Two years ago, processing a million tokens through a frontier AI model cost $33. Today, the same work costs 9 cents. That 99.7% price collapse is reshaping how companies buy, budget, and think about AI - and it's pushing tokens toward becoming something closer to a corporate currency than a billing metric.

Policy Notable Mar 21

Hachette Cancels Horror Novel 'Shy Girl' Over AI Writing Allegations

4,900 Goodreads ratings, a Big Five publishing deal, and a viral horror following. Then, on March 19, Hachette pulled the plug on Mia Ballard's Shy Girl after the New York Times brought evidence of AI involvement in the manuscript. It's the first time a major publisher has proactively cancelled a book over AI allegations, and it won't be the last.

Policy Notable Mar 21

Man Used AI to Generate Fake Songs, Stole $10M From Spotify and Apple Music

A North Carolina man just became the first person convicted in a federal AI-assisted music streaming fraud case, and the numbers are staggering: hundreds of thousands of fake songs, billions of fake streams, and over $10 million in stolen royalties.

Policy Notable Mar 21

Courts Are Demanding AI Reasoning Logs. Most Companies Can't Produce Them.

A company deploys an LLM to make consequential decisions. A court asks: show us how the system reached that conclusion. The company has nothing - no reasoning logs, no decision trace, no audit trail. The price tag: $10 million.

Open Source Mar 21

Daniel Miessler's 'Personal AI Infrastructure' Rethinks What AI Assistants Should Know About You

Your AI assistant forgets everything about you between sessions. Daniel Miessler thinks that's the core problem holding back the entire category.

Companies Notable Mar 21

Tanda's CEO: AI Coding Destroys the Core Constraint That Shaped Software Companies

$1,000 per day. That's what Tanda, an Australian workforce management company, budgets for AI coding tools - and their CEO Alex Ghiculescu says the number hasn't been hit yet.