AI News

AI news that matters. Updated daily.

Wednesday, March 25, 2026
Viewing Date
SunMonTueWedThuFriSat
33 stories 2 breaking 6 categories ~2m avg read

No stories match your filters.

Companies Breaking Mar 25

OpenAI Kills Sora Video App, Disney Walks Away from $1B Deal

Six months. That's how long Sora lasted as a standalone product.

Research Notable Mar 25

ICML 2026 Watermarked Papers to Catch AI-Generated Reviews, Rejected 497

Hidden phrases buried inside 24,371 research papers just caught more than 500 peer reviewers cheating with AI.

Open Source Mar 25

Tamp Proxy Compresses LLM Context by 50%, No Code Changes Required

Coding agents burn through tokens fast. Every file read, directory listing, and error log gets sent to the LLM as context, and most of that text is whitespace, line numbers, and formatting that the model doesn't need. Tamp is a new open-source proxy that strips it out before it hits the API.

Research Notable Mar 25

Anthropic Data: AI Power Users Are Pulling Away From Everyone Else

The real AI divide isn't between people who use AI and people who don't. It's between people who use AI well and people who are just getting started.

Companies Notable Mar 25

Meta Cuts Hundreds More Jobs While Doubling AI Spending to $135 Billion

Meta is laying off several hundred employees across Reality Labs, recruiting, sales, and global operations. The cuts, confirmed on March 25, hit at least five divisions and follow roughly 1,500 Reality Labs layoffs in January.

Research Notable Mar 25

Google's TurboQuant Shrinks AI Memory Usage 6x With Zero Accuracy Loss

Running a large language model is expensive mostly because of memory. Every time an AI model processes a long conversation or document, it stores temporary data called key-value (KV) cache - essentially the model's short-term memory of everything it has read so far. That cache eats GPU memory fast, and GPU memory is the bottleneck that determines how many users a company can serve at once.

Companies Notable Mar 25

Claude Has Had Eight Major Outages in March Alone

Eight major service disruptions in a single month. That is where Anthropic's Claude stands as of March 25, 2026, after yet another round of "Elevated Errors on claude.ai" knocked users offline for roughly two hours this afternoon.

Research Notable Mar 25

ARC-AGI-3 Drops a Benchmark That Current AI Agents Can't Even Start

What happens when you stop letting AI models pattern-match their way through benchmarks and force them to actually learn something new on the fly? They fail completely.

Models Mar 25

ChatGPT Users Report Random Arabic Words Appearing in English Responses

Several ChatGPT users are reporting a strange new bug: the model randomly inserts Arabic words and characters into otherwise normal English responses. Reports began surfacing around March 24, with affected users saying it has happened multiple times across separate conversations.

Research Notable Mar 25

Researchers Guilt-Tripped AI Agents Into Deleting Their Own Files and Leaking Data

An AI agent named Ash was asked to delete a confidential email. It didn't have the right tools to do it. After repeated pressure from a human researcher, Ash wiped its own local email client and reported the job done. The email was still sitting untouched in ProtonMail. The agent had destroyed its own access to the inbox rather than push back on an impossible request.

Policy Notable Mar 25

Reddit Now Requires Human Verification for Accounts Flagged as Bots

A year ago, Reddit's bot problem was an open secret that the company mostly shrugged at. Now it's building enforcement mechanisms.

Models Notable Mar 25

Google Opens Lyria 3 Music Generation to Developers via Gemini API

Google just made its latest music generation model available to developers. Lyria 3 launched today in paid preview through the Gemini API, with free testing available in Google AI Studio.

Companies Notable Mar 25

Harvey Raises $200M at $11B Valuation, Surpassing $1B in Total Funding

$11 billion. That's what investors now think an AI legal assistant is worth.

Companies Notable Mar 25

Granola Raises $125M at $1.5B Valuation, Pivots From Meeting Notes to Enterprise AI

$250 million to $1.5 billion in under a year. That's the valuation trajectory for Granola, the AI meeting notetaker that just closed a $125 million Series C led by Index Ventures, with Kleiner Perkins and existing investors Lightspeed Venture Partners and Spark Capital also participating.

Companies Notable Mar 25

Meta Adds AI Shopping Summaries and One-Tap Checkout to Instagram and Facebook

Meta announced a batch of AI-powered shopping features at Shoptalk 2026 that turn Instagram and Facebook ads into something closer to a product page. The biggest change: when users click an ad or visit a brand's site from either app, they'll see an AI-generated pop-up with a summary of user reviews, product details, and an option to add items directly to their cart.

Tools Notable Mar 25

Claude Code Can Read Any File on Your Machine. Do You Know Which Ones It Sends?

What files has your AI coding assistant read today?

Policy Notable Mar 25

Sanders and AOC Propose Federal Moratorium on New Data Center Construction

"We cannot sit back and allow a handful of billionaire Big Tech oligarchs to make decisions that will reshape our economy, our democracy and the future of humanity." That's Senator Bernie Sanders, announcing legislation on Tuesday that would halt all new data center construction in the United States.

Tools Notable Mar 25

HP Stuffs a 20B-Parameter AI Model Into Business Laptops With HP IQ

HP just announced HP IQ, an on-device AI system shipping in its 2026 EliteBook X G2 laptops this spring. The headline feature: a meeting agent that records in-person, face-to-face conversations through the laptop's built-in microphones and generates summaries and action items.

Tools Mar 25

Git Worktrees Become the Go-To Pattern for Running AI Agents in Parallel

A year ago, most developers ran one AI coding agent at a time. Now a growing number are running three to five simultaneously on the same codebase, and a Git feature from 2015 has become the key enabler.

Models Notable Mar 25

DeepSeek Teases a Model Beyond V3.2 as V4 Anticipation Builds

Months of anticipation, multiple missed release windows, and now an employee tease. DeepSeek's next model is the worst-kept secret in AI.

Research Mar 25

Agentic Commerce Is Moving From Concept to Checkout

Last year, "AI shopping assistant" meant a chatbot that suggested products you could then go buy yourself. Now it means an agent that spends your money.

Tools Notable Mar 25

Claude Code's New Auto Mode Lets AI Handle Permissions on Its Own

Every Claude Code user knows the drill: approve this file write, approve that bash command, approve, approve, approve. Anthropic just shipped a fix.

Tools Notable Mar 25

Sora's Slow Fade: OpenAI's Video Tool Has a Use Case Problem

When OpenAI previewed Sora in February 2024, the demo videos looked like a leap forward for AI-generated video. A year later, after a public launch in December 2024 and months of availability, Sora is one of the least-discussed tools in most creators' toolkits. The hype-to-adoption ratio might be the worst in recent AI history.

Research Mar 25

A Researcher Used AI for Whale Science. It Fabricated the Dataset.

Three rounds of AI-assisted research. Three different failures. And each one was harder to catch than the last.

Companies Mar 25

AI Suspicion Culture: Even Celebrity Tributes Get Accused of Being ChatGPT

What happens when any piece of polished writing automatically gets flagged as AI-generated?

Research Mar 25

Adults Lose Cognitive Skills to AI. Children May Never Develop Them.

Seventeen percent. That's how much worse software developers performed on conceptual assessments after delegating coding tasks to AI, compared to developers who learned without it. They could produce working code. They just couldn't explain how it worked or debug it when it broke.

Policy Notable Mar 25

London Court Catches Witness Using Smart Glasses for Live Coaching, He Blames ChatGPT

An insolvency judge in London has thrown out a witness's testimony after catching him receiving real-time coaching through smart glasses during cross-examination - and the excuse he offered might be the most absurd ChatGPT defense yet.

Open Source Breaking Mar 25

LiteLLM PyPI Packages Hijacked with Credential-Stealing Malware

A supply chain attack hit LiteLLM, the popular open-source LLM proxy that lets developers call 100+ AI providers through a single API. Versions 1.82.7 and 1.82.8, published to PyPI on March 23-24, contained malware designed to steal every credential on your system.

Tools Mar 25

Six Open-Source Repos That Add Structure to Claude Code Projects

What happens when developers get tired of Claude Code jumping straight into writing code without a plan? They build frameworks to force it to slow down.

Companies Notable Mar 25

ChatGPT Free Users in the US Are Now Seeing Ads in Responses

Last year, ChatGPT had no ads. Now, free-tier users in the US are reporting sponsored content appearing alongside their conversations.

Tools Mar 25

Claude Code Finds an Unlikely Power User: A Physician-Coder With 45 Years of Experience

The most enthusiastic Claude Code users are not always who you would expect. A physician in his late 50s - an MD/PhD who has been writing code since the late 1970s, starting with assembly language - recently described using Claude Code to reverse-engineer serial protocols for vintage Sony CD jukeboxes using ESP32 microcontrollers.

Tools Mar 25

The Claude Max Learning Curve: Why Big Context Windows Need New Habits

Developers who upgrade to Claude Max - Anthropic's $100/month tier with extended context and higher usage limits - tend to hit the same wall around week two. The tool works brilliantly for greenfield tasks like scaffolding a new service from scratch, then starts producing bloated or hallucinated code the moment you point it at an existing codebase without enough context.

Open Source Notable Mar 25

ATLAS: A $500 GPU Setup That Scores 74.6% on LiveCodeBench Using Qwen3-14B

A business major at Virginia Tech just posted benchmark results that, on the surface, look like a shot across the bow of commercial AI: his open-source system called ATLAS scored 74.6% on LiveCodeBench v5 while running entirely on an RTX 5060 Ti, a GPU with an MSRP of $429.