Short answer: rescue the app when the data model is sound and the problems are keys, access rules, login, deployment or a handful of screens. Rebuild only the layer that is broken when every fix breaks something else, or when the data model itself is wrong. Rebuilding everything is rarely the only o
AI News
Aggregated from 20+ sources every 8 hours — official AI blogs, tech media, and research feeds.
Last updated: Nov 22, 2026
50 articles
The AI for Business series follows Alex, a fictional operations manager at a small company, as a running example of putting AI tools to work day to day. What happens to text you paste into an AI tool at work depends on three things: whether the service may use it to train its models, how long it kee
Hey, Your designs work when a request can wait a second. Then a chat message, a shared document edit or a video call has to land now, and the same design falls over. Course III is for that gap. Try this today, free. Pick a feature in your app that notifies people, a comment alert or a chat message.
TL;DR Meta's Muse personal agent runs on a dedicated Secure VM, with a separate Sentinel agent approving every internet-facing action and a full audit trail visible to the user. Credentials go into secure storage so Muse can use them without seeing them; a Confidential VM with user-held keys is plan
Don't have a GPU for fine-tuning? Google Colab gives you a free one, and it's enough to fine-tune an LLM. In this quick walkthrough, I show you how to set up Unsloth Studio and run it inside a Colab session using a Cloudflare Tunnel. 📓 Colab notebook: Unsloth Studio – Google Colab 🎥 Want a deeper
Letting an agent query the database is the fastest way to stop it guessing. "Does the migration backfill status?" is a 1-second SELECT, not a 10-minute code read. It's also the fastest way to have a bad afternoon. These are the guardrails I put in the SignOff Studio MCP server, and why each one is e
If you use a coding agent for more than an afternoon, you already know the tax: the model does not remember which file you had open, which doc you skimmed, or which page you read twenty minutes ago unless you paste that context back in. I built Ambient Memory to remove that tax without sending your
x402 Protocol: Machine-Native Micro-Settlements for AI Agents Web HTTP memiliki kode status yang terbengkalai sejak 1991: 402 Payment Required. Tim perancang awal membayangkan web memiliki rel pembayaran terintegrasi, tetapi perbankan tradisional tidak mampu memproses transaksi sub-sen tanpa biaya k
METR reports that leading AI models like Opus 5.5 and GPT-6 Astra are improving rapidly at research tasks but getting harder to measure. Current data can't rule out a dangerous acceleration in AI development, the organization warns, though it also says common tests may overestimate what the models c
This is The Stepback, a weekly newsletter breaking down one essential story from the tech world. For more on the vibe-hacking of everything, follow David Pierce. The Stepback arrives in our subscribers' inboxes on Sunday at 8AM ET. Opt in for The Stepback here. How it started Like so many things wit
Security researchers at Dragos used two million fake AI requests to uncover serious vulnerabilities in the AI supply chain, finding at least three compromised providers, hundreds of stolen AWS keys, and marketplace resellers who were simply reading customer requests. OpenRouter responded by removing
Reuters reported on November 22, 2026 that Anthropic limited deal teams to four senior bankers at some of its 22 IPO underwriters and asked some bank officers to accept six-month restrictions on work with AI rivals, as part of a tightly controlled listing process. Several people close to the offerin
The Lean and Mathlib community has officially confirmed Levent Alpöge's proof of the Riemann hypothesis as correct. Dozens of independent verifiers checked the 17 million lines of computer code. The solution arrived years earlier than expected: Terence Tao had predicted it wouldn't happen until arou
Major US financial institutions are banning third-party bots from logging into your accounts—and warning that any costly bot mistakes are strictly your problem.
Hermes Agent, OpenClaw, Gemini Spark, OpenAI dots, Meta Muse and Perplexity Personal Computer ranked on memory, security defaults and real cost. The post Best Personal AI Agents in 2026: OpenClaw, Hermes Agent, Gemini Spark and More, Compared on Memory, Security and Cost appeared first on MarkTechPo
Google launched Gemini API Tool Retrieval in preview. Agents keep a large tool library on standby and load each parameter schema only when needed, inside one Interactions API request. The post Google AI Previews Tool Retrieval in the Gemini API for Large Tool Libraries appeared first on MarkTechPost
OpenAI's GPT-6.2 Astra and Sol give much shorter, more direct answers than their predecessors, but they are also less willing to admit uncertainty. Anthropic's Claude models score well ahead on transparency. The article GPT-6.2 Astra and Sol get to the point faster but score lower on transparency th
Tesla announced on November 21, 2026 that Full Self-Driving (FSD) v15 was starting to roll out, with an upgraded driving network, a live Tesla Reality View visualization and a company-reported 25-fold improvement in its critical-intervention rate. Navigation, parking, driver monitoring and speed han
OpenRouter announced on November 22, 2026, that it had off-boarded three inference providers and removed 125 endpoints from eligibility for Zero Data Retention (ZDR), a setting under which prompts and responses are not stored. It also notified all potentially affected customers. The findings did not
TensorLean, from Columbia University and CertiK, formally verifies production GPU kernels at the PTX level using Lean. It checked 149 kernel instances, proved 83 correct, and found 38 bugs in FlashAttention, DeepGEMM and more. The post A Team of Researchers from Columbia and CertiK Introduce TensorL
Together AI disclosed on November 21, 2026 that an unauthorized party downloaded stored samples from a subset of customer prompts submitted to its serverless inference service. An AI coding agent had exposed the private-repository credential used in the download while the company was developing its
Signal is limiting users' ability to take screenshots in both its Android and iOS apps.
The Lean and Mathlib communities certified on November 21, 2026 that Levent Alpöge’s 17-million-line Lean library correctly states and proves the Riemann Hypothesis, following a day of compilation, statement-matching, and proof-kernel checks. Their joint statement was endorsed by the Lean FRO and th
We got two very different pictures of Nvidia — the company’s latest earnings report vs. the tech executives loudly calling for Jensen Huang’s return.
One of the items on the agenda for UN diplomats meeting in Geneva this week was working towards rules regarding the use of autonomous weapons, or so-called "killer robots." Eventually, representatives left the meetings agreeing to "start discussing options" after a largely fruitless week of negotiat
We created a list of the most notable AI agents that can live in your text messages, from general assistants to agents designed for families, travel, and work.
Nvidia researchers argue that language models often don't need their full context windows—which are slow, expensive, and prone to errors—and propose an alternative. In tests, periodic summaries and targeted keyword search usually matched full-context results at a fraction of the cost. The team even
A Buddhist monastery in Vermont called Maple wants to establish itself as Silicon Valley's spiritual authority while creating its own AI god. According to a New Yorker report, leader Soryu Forall's group records every breath its members take to train an AI in his image. Former members report manipul
An actor has licensed his face to a microdrama startup — in a trend that's already popular in China. Experts question the long-term consequences if this becomes a trend in Hollywood, too.
Today, I’m talking with Hayden Field, The Verge’s senior AI reporter, about the Anthropic IPO for the third and final episode in our series about AI in real life. The AI lab behind Claude went public on the Nasdaq last Thursday. The numbers are gigantic, but not really that important for this conver
Chess.com is ready to go a lot bigger than chess. After launching Gambit, a gambling-free poker site, earlier this year, the company is now adding 10 more games: backgammon, checkers, go, hearts, mahjong, shogi, sudoku, solitaire, xiangqi, and a Scrabble alternative called word. They're all part of
A person receives an automated decision that makes no sense to them. They ask for a review. Somewhere inside the organization, a policy names an accountable executive—but the support team cannot find anyone authorized to change the result. Governance exists on paper. The person is still stuck. That
CNBC reported on November 21, 2026 that Anthropic shares closed their second trading session on November 20, 2026 at $88.91, down 8.5%, as the Nasdaq Composite completed a 10% weekly loss, its worst week since the early Covid market turmoil in 2020. The close left the Claude developer with a fully d
OpenAI has released new prompt templates for GPT-6.2 Astra and Sol meant to fix common model behaviors such as overly brief answers, unnecessary follow-up questions, and excessive testing when coding. The API is changing as well, since editing a conversation's history can discard reasoning content a
AI is forcing companies to patch their systems faster as attackers race to exploit newly discovered vulnerabilities. Welcome to internet outage hell.
Thinking Machines released Inkling 2 Mini, a 26.4B-parameter looped MoE transformer with 3.4B active parameters. It runs on a 32GB Mac or one DGX Spark and scores 79.2% on SWE-bench Verified. The post Meet Thinking Machines Inkling 2 Mini: 26B Looped MoE Runs on a 32GB Mac appeared first on MarkTech
OpenAI has released GPT-Live-1.1 in the API. The full-duplex voice model improves multilingual accents and transcription, and a new standby mode bills idle WebRTC time at $0.02 per minute instead of $0.05. The post OpenAI Ships GPT-Live-1.1 for Voice Agents, Cuts Idle Cost to $0.02/Min appeared firs
Anthropic mathematician Levent Alpöge says he used Claude to prove the Riemann hypothesis, math's most famous open problem for 167 years, with a proof formalized in Lean that spans over 250,000 pages. Just hours later, Stanford researchers published an independent, entirely human competing proof, an
Google announced on November 21, 2026 that it is providing the Indian Institute of Technology Kharagpur (IIT Kharagpur) with $5 million to support advanced AI research, expand AI education and build AI capacity across the university’s campus. The funding is designed for university researchers and st
Baseten released bison-1 and groundhog-1, open-weight decision models built on Gemma 4. bison-1 scores 76 on the AA Decision Index v2.1, and both models support 128k context and image input. The post Baseten Releases Bison-1 and Groundhog-1, Topping Decision Model Index appeared first on MarkTechPos
The blockbuster game makes no secret of its contempt for law enforcement. Real police departments using it to recruit new members and scold the public apparently didn’t get the memo.
Test using prompts with unique, trackable secrets proves that data doesn't stay private.
Spotify founder Daniel Ek is backing Europa Fusion, a fusion power startup founded by an AI researcher. Here's what the startup is building.
Amid allegations of multiple affairs, including with a direct report and an alleged Chinese intelligence asset, sources tell WIRED the Nvidia board’s main issue was the former CEO’s “lack of candor.”
Mathematicians are having their minds blown by the latest AI-driven result — and there's a competing human solution, too.
President Donald Trump promised to put a "tremendous tax" on companies that build data centers outside of the US, even as states and cities around the country are placing restrictions on new applications to build them. "We're not going to allow our companies to build data centers outside of our coun
A new report from the AI evaluation nonprofit METR finds that some AI capabilities are continuing to improve along an exponential trajectory—but it also indicates that models aren’t quite as good at tasks that require judgment as they are at coding. The report is an update to the organization’s time
Google Connected Apps simplify your year-end tasks. Manage groceries, credit scores, and home services directly in Search. Get the details for November 2026.
RadixArk released DaisyRose-27B, an Apache 2.0 agentic coding model post-trained from Qwen3.8-27B with reinforcement learning only. It more than doubles Terminal Bench 4.1 pass@1 to 12.0 and runs on a single GPU. The post RadixArk Ships DaisyRose-27B, a 27B Agentic Coding Model Trained With Reinforc
Lenders don't want SpaceX shooting their collateral into orbit. What does that mean for space data centers writ large?