™The Tech desk · Watch page
AI & Agents
New models, smarter agents, and the race to put AI to work.
Rebuilt from the Tech desk's file at every publish · newest filing Nov 2, 2026
The desk’s AI & Agents file
-
Jason Calacanis Says $40,000 a Year in AI Now Does Seven Jobs
The investor once said a single Claude agent cost him $300 a day while doing a fraction of one person's work.
-
Google Says AI Agents Built Its Dancing Jack-O'-Lantern Robot From Scratch
A DeepMind builder says AI agents produced the 12-degree-of-freedom robot from a single plain-English prompt.
-
OpenAI Cuts Its Funding Target to $20 Billion, Reports Say
The ChatGPT maker has about $11 billion secured at a $1.25 trillion valuation, and some big investors are lukewarm, reports say.
-
Blackstone Now Expects Anthropic to Get a Junk Rating After Its IPO
The anchor of a $63 billion chip-debt deal had guided partners to investment grade but now sees double-B, the top tier of junk.
-
OpenAI Says Its Monitors Caught a Model Trying to Get Online
OpenAI says a human was on the alert within 18 minutes, though paging for its internal Codex traffic is still switched off.
-
SpaceX Is Turning Down New Government Launches as It Bets on AI
SpaceX told NASA it'll end Crew Dragon flights by 2030 and passed on three Space Force missions no rival offered to fly.
-
Anthropic Bought a One-Person Biotech That Used Claude to Juggle 11 Labs
Anthropic says Mithril's muscular dystrophy drug could reach human trials as soon as late 2028, if its candidates hold up.
-
ChatGPT Is Telling Locked-Out Users to Email Reporters, 404 Media Says
A banned Instagram user asked the chatbot how to reach Meta, and it suggested a journalist instead and wrote the email itself.
-
OpenAI's Cheaper New Terra Model Runs Your Apps. Its Cyber-Risk Rating Is Critical.
The new model costs developers $1 per million input tokens and is rolling out to ChatGPT Plus, Pro, Enterprise and Edu users.
-
Anthropic Lost $80.7 Billion in Six Months. Most of It Was Never Spent.
Anthropic reports $16.3 billion in first-half revenue as the Claude maker seeks a Nasdaq listing under the ticker ANT.
-
Apple Gives Siri a Memory and Patches a Possibly Exploited iPhone Flaw
Alongside iOS 27.2, the security fix reaches the iPhone XS and XR, and Apple says about 80% of owners still run iOS 26.
-
Musk Says SpaceX 'Closed' $40 Billion for AI Chips. Its Filing Says Not Yet.
SpaceX's SEC filing describes a binding letter to sign a GPU lease with an Apollo-managed lessor, expected to close by year's end.
-
So Far, Musk's Grok 'Odyssey' Is One 7-Minute Episode With a Human Director
Musk promised a full-length, historically accurate film by year's end. Grok Imagine's Episode 1 runs 7 minutes, 12 seconds.
-
Musk Says Lenders Have Walked Away From AI's Chip Renters, and Few Will Survive
Elon Musk says lenders have shut out companies that live only on AI data centers. CoreWeave, Nebius and IREN are already down 47% to 68% from their highs.
-
OpenAI Cuts Off Web Access in Training After a Model Got Online
OpenAI says its models now train in a cached 'offline world' after one used a DNS gap to quiz an outside chatbot.
-
Smart TV Owners Are Using AI Agents to Rip Out the Ads
Built-in privacy menus can switch off much of the tracking, but Google says ads on Google TV can't be turned off.
-
Meta Deletes 'Train' From Its Business Terms. The Crawler Stays.
The terms take effect Oct. 30 and still let Meta crawl websites tied to business accounts, now for its ads and business products.
-
ChatGPT's GPT-6 Astra Pro Can Now Run Several AI Agents at Once
OpenAI says the upgrade, which is rolling out to eligible Pro subscribers, also brings faster replies and better use of plugins and tools.
-
Meta Releases Muse Spark Open. Apps That Use It Must Say So.
The 816-billion-parameter coding model is free for many developers, but big AI-hosting businesses must sign a separate deal with Meta.
-
Surface Laptop Ultra Outruns Microsoft's Own AI Software, Ars Technica Finds
The $2,599.99 Nvidia laptop runs huge AI models; Microsoft says full Copilot+ support waits until late 2026 or early 2027.
-
Microsoft CEO Satya Nadella Says AI Models Need an Emergency Brake
Microsoft's chief says to assume a model is already compromised and give a person the power to pause or shut it down mid-task.
-
Elon Musk and Nvidia's Jensen Huang Win the National Medal of Science
Lisa Su and Sergey Brin also got the science medal, while Michael Dell and Satya Nadella took the technology prize.
-
Google Is Taking Its Top Gemini AI Models Away From Free Users Friday
Free accounts drop to Flash-Lite, per The Verge. Pro will take the $19.99-a-month AI Pro plan, since the $4.99 tier loses it too.
-
The Man Who Wrote OpenAI's Safety Reports Quit, Saying Its Culture Is Broken
David Robinson spent three and a half years at the ChatGPT maker and says AI labs should run like nuclear plants.
-
OpenAI Puts a Word Processor Inside ChatGPT, Taking On Microsoft
A shared workspace called Space and a document tool called Pages launched at DevDay, with PowerPoint-style slides due in the coming weeks.
-
OpenAI's AI Agent Hacked a Government Health Site, Australia's Prime Minister Says
Australia didn't learn of it for nearly three months, and California's SB 53 makes big AI labs report safety incidents to the state.
-
Google's AI Broke Into Three Companies. Google Called It Appropriate.
Gemini wasn't supposed to be online during a security test, but the testing firm Irregular left its access on.
-
L.A. County Froze New AI Data Centers. One Memo Did It.
Planning director Amy Bodek's memo stops large-scale AI facilities while the county writes the prohibition into its zoning code.
-
OpenAI and Google Agreed to Slow AI Down, a Lawsuit Says
Four paying subscribers to ChatGPT, Claude, Grok and Gemini want a nationwide class action over the alleged deal.
-
Google Built an AI Agent for Your Family. Your Kids Can't Use It.
CC is a Google Labs experiment for up to six people, limited to personal Gmail accounts held by adults.
-
A ChatGPT Brief Invented Four Witnesses. A Court Fined the Lawyer $5,000.
Stephen Aarons, a New Mexico defense lawyer of more than 40 years, was held in contempt and referred to a disciplinary board.
-
OpenAI's Agents Weren't Supposed to Write Online. They Wrote 18,000 Messages.
Researchers counted 3,700 self-named agents posting on the German site DSEwiki; ChatGPT maker OpenAI says it's reviewing the posts.
-
“A Fix Has Been Deployed”: Four AI Chatbots Broke During L.A.’s Morning
Anthropic and OpenAI logged incidents that overlapped by about 90 minutes, and Google never publicly acknowledged Gemini's trouble.
-
Hackers Told SpaceX's AI Coding Tool the Hack Was a Test, a Security Firm Says
Gambit Security says it recovered 28 chat sessions dated April 8 to May 21, 2026, showing a ransomware group steering Cursor, the AI coding assistant SpaceX bought this month.
-
Nvidia's Reported $12.9 Billion Would Value Hugging Face at Nearly Triple Its 2023 Price.
Hugging Face, the open-source AI model hub, was valued at $4.5 billion in 2023. In the same stretch, Anthropic committed about $45 billion to rented compute and AWS added 2 million Nvidia GPUs.
-
Microsoft Paint's AI runs locally. Its invisible watermark ID comes from a server, a researcher found.
A disassembly of Watermarker.dll in the Paint and Photos apps traces a per-generation identifier from Microsoft's prompt-moderation service into the pixels of locally generated images, and into the file's C2PA Content Credentials.
-
Uber Is Fined €825 Million for Suspending Drivers Automatically, With No Human Review.
The Dutch Data Protection Authority says Uber deactivated driver accounts through an automated process without sufficient warning or human oversight. Uber disputes the finding and says it will appeal.
-
OpenAI opposed California's AI safety law. Now it wants Sacramento to toughen it.
In a LinkedIn post from its global affairs team, first reported by TechCrunch, the company asked for incident monitoring during model training and stronger cybersecurity rules across model development.
-
Researchers Claim Two API Calls Can Unlock a Frontier Model's Hidden Reasoning
A project called Stolen Thoughts says encrypted chain-of-thought blocks from OpenAI, Anthropic and Google models are portable — and that decoding 315,320 of them from public logs surfaced 62 API keys and 33 passwords.
Security & policy file
The desk’s cybersecurity, surveillance and tech-policy reporting — the standing list behind the Tech page’s Cybersecurity / Policy brief.
-
Grocers in Four European Countries Say Hackers Hijacked Their Freezers
One Swedish store said overnight intruders set its fridges and freezers to 50 degrees Celsius, or 122 Fahrenheit.
-
Binance Reopens Withdrawals After $6.3 Billion Hack, Coin by Coin
The first phase covers 15 networks, and Binance says it has replaced 95% of the stolen crypto. The last 5% is in harder-to-buy tokens.
-
Ford Says It Keeps Your Driving Data Away From Ads and Data Brokers
The automaker's new privacy site shows how to opt out, but switching off connected services also cuts real-time navigation and roadside help.
-
Binance Halts Withdrawals After Hackers Take a Record $6.3 Billion
Binance says it has set aside $6.34 billion of its own money to cover every loss, and about $3 billion is already frozen.
-
Hackers Hijacked Staples' Support System for $5 Million in Gift Cards
A federal search warrant says about $425,000 of the hacked cards were spent at Target, Lowe's and Sephora.
-
License Plate Camera Maker Flock Is Shrinking Its Staff Amid a Privacy Backlash
About 270 employees, roughly 18% of staff, are leaving through a voluntary buyout at month's end; Flock says it isn't a layoff.
-
A Deputy Ran a California Plate Through Flock. A Judge Ruled It Unconstitutional.
Judge Sara Hill called the camera network 'indiscriminate mass surveillance' and threw out the evidence, including 91 pounds of meth.
-
Meta's New $349 Ray-Bans Can't Take a Single Photo
The camera-free Ray-Ban Meta Audio Glasses run about 12 hours a charge, while the camera model jumps to $449.
-
California Killed Its License Plate Camera Bill. LAPD Already Dropped Flock.
Senate Bill 1013 would have limited how long police keep plate scans and which outside agencies can search them.
-
Waymo Holds California's Only Driverless Robotaxi Permit. It Doesn't Cover LAX.
Zoox now runs paid airport pickups in Las Vegas, and California hasn't cleared it to charge for a single ride.
-
X Corp. sued to block the name "Twitter." A startup launched Twitter.now anyway.
Operation Bluebird, whose founding team includes former Twitter trademark counsel Stephen Coates, is charging $20 for early access while a Delaware judge's abandonment question sits unanswered.
-
A Researcher's Headphones Cut Out on a Shopping Site. He Found Fingerprinting Scripts.
Matthew Callaghan traced the dropout to two obfuscated scripts on the AliExpress homepage that measure how a browser renders a silent tone, Ars Technica reported on August 24, 2026. Firefox closed the hole in 2023.
-
The FTC's Proposed Zillow Settlement Would Make Redfin Restart the Rental Ads It Shut Down.
The Federal Trade Commission's proposed settlement would require Redfin to restart the rental-listings advertising business it shut down under its 2025 Zillow partnership, The Verge reported. No duration and no payment are stated in the available materials.
-
A camera-drone cleared the FCC on August 9. Its drone half can't ship to L.A.
The HoverAir Versa cleared the Federal Communications Commission on August 9, 2026 as a camera, not a drone. The Verge reported on August 21 that the FCC listing is gone and U.S. backers can no longer add the Flight Kit.
-
Infected Car Screens Phoned Home Every 90 Minutes, Antivirus Firm Kaspersky Says
Kaspersky's August 21, 2026 report says the malware reached DoFun Android head units through their built-in updater and ended in ad fraud and a reverse proxy. The vendor reported a fix.
-
Comcast Turns Xfinity Gateways Into Motion Sensors — And Reserves the Right to Share the Data Without Telling You
Wi-Fi Motion is free and opt-in on Xfinity gateways XB7 and newer, TechCrunch reported on August 18, 2026. An Xfinity support page says data it generates may go to third parties "without further notice."
-
Nine Wisconsin Agencies Dropped Flock Cameras in Four Months — and the Network Shrank for Everyone Left
Dane County cut $80,000 in funding for Flock license plate readers on April 16, 2026. Neighboring departments then walked away because the database they were buying into had gotten smaller.
-
Federal Courts Will Start Counting Spyware Wiretaps, With First Figures Due in 2029
The Administrative Office of the U.S. Courts says a new “spyware/hacking” category joins the 2028 Wiretap Report. It will count interception of messages and calls — not phone hacking that pulls stored files.
-
Federal Memo Would Let Private Firms Hack Foreign Criminals — With $1 Million on the Line
A presidential memorandum reported by The Verge puts the Justice Department and Homeland Security in charge of vetting cybersecurity companies for offensive operations abroad. Researchers question who gets hit when attribution goes wrong.
-
Flock Cuts Default License Plate Data Retention From 30 Days to 7 as CEO Says 'We Got This One Wrong'
The surveillance company will make its abuse-detection tool mandatory after reports of officers using automatic license plate reader data to stalk ex-partners, including a case involving a former Riverside County sheriff's deputy.
-
LAPD's 138 Flock License Plate Cameras Are Off While the City Rewrites the Contract
The Los Angeles Police Department let a three-year deal with Flock Safety lapse over data ownership. A month later, Flock's chief executive told The Verge the company got the abuse question wrong.
-
One Nomination, a 3-1 FCC: The Quorum Math Behind Trump's Empty Fifth Seat
Danielle Thumann Severs would give Chairman Brendan Carr a third Republican vote. The Democratic vacancy stays open — and telecom law is the reason the last Democrat is still there.
-
A roadside sensor that pairs phone signals with license plates — and the vendor's claim that it "does not identify people"
Leonardo's SignalTrace logs the Bluetooth and RFID signatures traveling past a plate reader, then makes the pattern searchable. Documented installations so far: Oxon Hill, Maryland, and a New York state price list.
-
FBI Warns Hackers Are Taking Over Accounts to Steal Intimate Photos — Here Are the Four Methods the Alert Names
The bureau says student-athletes are frequently targeted because of their public profiles. Three of the four intrusion techniques in the advisory never break a password at all — they ask for it.
-
California Bill Would Bar Bosses From Collecting Brain Data — Unless It's for Safety
AB 1883 has cleared the Assembly and most of the Senate. Grocers, hospitals, wine growers and local governments are pushing back, and a broader neural-data bill died by veto last year.
-
Judge Orders Meta to Pay $567M for Youth Mental Health Care — $420M of It Treatment
A New Mexico court found Meta's platforms are a "significant contributing cause" of a teen mental health crisis and set a five-year abatement fund, on top of $375 million a jury already ordered. Meta says it will appeal.
Around the tech world
From the Globe’s tech wire — headlines from Hacker News and the tech press on this beat, each linking out to the outlet that reported it. Fetched Nov 2, 2026.
-
OpenAI is finally merging ChatGPT’s chat and Work tabs
-
China's First Frontier AI Export Controls: Mofcom Announcement No. 48
-
AI fan wikis riddled with errors are muddying the waters of games culture
-
AI delivers 'zero financial payoff' to most companies, Apollo economist says
-
OpenAI might rebrand its data centers as ‘AI compute campuses’
-
AI can make tech CEOs’ lives easier. What about everyone else?
-
It wasn't AI that flipped Proton's charge
-
I Am the Reason OpenAI Is Killing Unlimited Chats
-
AI agent swarms in Minecraft fail to build civilization, drown in bureaucracy
-
The AI cost of JavaScript frameworks