01
Engadget ↗
OpenAI is testing Dots, its always-on personal agents, with a small number of people on the $20 a month ChatGPT Plus plan. The agents are also opening to every ChatGPT Business Standard user. Dots can work without a person watching, connect to a computer, and reach more than 4,000 apps. They began on GPT-6 Astra and can now run on GPT-6.1 Terra as well.
02
Anthropic ↗
Anthropic said Claude Fable 5.2 finished two open calculations in black hole theory. One shows a sharp switch from hot gas to a black hole still appears where spacetime has no definite shape. The other proves an exact ceiling, about 9.13, on the lightest black hole in a model universe. The results were checked in Lean in September, and they do not describe our universe.
03
Google ↗
Google booked Blue Origin's New Glenn rocket for the first heavy-lift launch of Project Suncatcher, targeted for the first quarter of 2029. The satellites, each carrying Google's own AI chips and about as much compute as one data center rack, will go into an orbit with near-constant sunlight. Google calls the project a long-term test of running machine learning on solar power in space. It follows a smaller prototype flown in October.
04
TechCrunch ↗
Meta employees wrote internally that the Muse agent mixed up conversations across accounts during testing, according to a Reuters review of those posts. Examples included childcare requests, orders, and bank balances that were not the employee's. One tester said Muse reminded her about a Signal message she had not given it access to. Meta said employee builds differ from the public app, which has more than 10 million downloads.
05
The Verge ↗
Signal is blocking screenshots in its iOS and Android apps so system agents such as Siri and Gemini cannot read messages on screen. Users can turn the block off, and Signal says doing so can expose chats to those agents. The block is not available on Mac. President Meredith Whittaker said operating-system makers do not give app developers a clean way to keep agents out.
06
The Verge ↗
Waymo plans to end its exclusive robotaxi deal with Uber in Austin and Atlanta in January 2028, so riders can hail cars in the Waymo app. Serve Robotics said it will not renew its Uber Eats partnership when that deal expires in 2027, and it is opening a direct interface for local shops. Uber said it had raised reliability concerns with Serve since early 2026. Uber still hopes to keep non-exclusive Waymo rides after 2028.
07
Ars Technica ↗
Four students sued after an Omnilert camera at a Baltimore County high school flagged a Doritos bag as a gun on October 20, 2025. Safety staff canceled the alert in about two minutes, but the principal still called police, and officers handcuffed the students. The suit asks a court to stop the district using Omnilert until staff and police are trained. Omnilert said the fear should not have happened and that a canceled alert must stay canceled.
08
BBC ↗
The Dutch data authority tested six smart glasses sold in the Netherlands and told people not to buy them. It did not name the brands. In the tests, one pair claimed an ambulance was coming when none had been called, and another gave dating advice about a child. Some recording lights could be covered with tape and the camera still worked, and on some models a nearby person could connect without permission.
09
Bloomberg ↗
SpaceX sold $32 billion of bonds on Thursday to fund a lease of Nvidia AI chips, in a deal meant to stay off its own balance sheet. The largest piece, about $25.2 billion, was priced to yield 8.259 percent. The bonds are part of a financing of more than $40 billion that also includes bank loans. Goldman Sachs and Apollo ran the sale, and the chips are leased for about five years.
10
Financial Times ↗
Germany's information-security office said attacks on political institutions and businesses have surged, and it raised the outlook to critical. Its cyber index moved from 130 to 238 since January, an 83 percent rise. Interior minister Alexander Dobrindt said the threat is high. The office said the average time from finding a new flaw to attackers using it has fallen to about eight days this year, from about two months at the end of 2024.