The Trump administration has agreed to work with Google DeepMind, Microsoft, and xAI to perform government safety checks on their new AI models before and after release. This marks a change from President Trump’s earlier rejection of Biden-era AI safety policies, prompted by concerns over the risks posed by advanced AI systems.
Key Facts
The Trump administration signed agreements with Google DeepMind, Microsoft, and xAI for AI safety testing.
Previously, President Trump had dismissed the need for voluntary AI safety checks and renamed the US AI Safety Institute to remove “safety” from its name.
Anthropic delayed releasing its advanced AI model called Claude Mythos due to fears it could be misused.
The Center for AI Standards and Innovation (CAISI) will conduct tests on AI models, sometimes with reduced safeguards to assess risks better.
A task force of experts from various government agencies has been formed to focus on AI national security concerns.
Some companies, like Google DeepMind and Microsoft, support the government’s AI safety testing plans.
Critics worry CAISI might not have enough resources or expertise and that voluntary agreements may fail to provide enough transparency.
There is concern that political bias in AI evaluations could reduce public trust and discourage companies from cooperating.
Read the Original
Want the full story? Tap a source to open the original
article.
Shivon Zilis, a former OpenAI board member, testified in a court case involving Elon Musk and OpenAI's shift from nonprofit to for-profit status. She revealed that Musk offered to donate sperm to her in 2020, leading to the birth of four children, and described her close working relationship with Musk across several of his companies, including OpenAI.
Key Facts
Shivon Zilis was an advisor and board director at OpenAI from 2016 to 2023.
Around late 2020, Elon Musk offered to donate sperm to Zilis, which she accepted.
Zilis and Musk have four children together.
Zilis had a brief romantic relationship with Musk about 10 years ago but was not romantically involved with him around 2020.
Musk encouraged those around him to have children at that time.
The paternity of the twins born in 2021 was initially kept confidential between Musk and Zilis.
Zilis told OpenAI CEO Sam Altman about Musk's paternity after a media report was imminent.
Zilis worked at Musk’s companies Tesla and Neuralink, besides her role at OpenAI.
The court case involves Musk's lawsuit to reverse OpenAI’s change to a for-profit company.
Emails presented in court show Musk wanted more control of OpenAI and suggested it join Tesla.
OpenAI founders preferred transitioning to a B Corporation, a type of company with a social mission and profit.
Read the Original
Want the full story? Tap a source to open the original
article.
Meta is being sued because it is accused of using writers’ copyrighted work without permission to train artificial intelligence (AI). The lawsuit claims Meta used authors’ stories and ideas to help AI generate similar content quickly.
Key Facts
Meta is facing a legal case about how it trains its AI systems.
The lawsuit says Meta used copyrighted material from writers without asking.
Writers spend many years creating stories and ideas.
AI can recreate stories and content in seconds after training.
The case raises questions about using creative work to teach machines.
Read the Original
Want the full story? Tap a source to open the original
article.
Apple has agreed to a $250 million settlement over a lawsuit claiming it misled customers about Siri’s features. Eligible iPhone owners who bought certain models between June 2024 and March 2025 could receive up to $95.
Key Facts
The lawsuit says Apple advertised "Enhanced Siri features" that were not yet available.
The settlement covers people who bought iPhone 16, iPhone 15 Pro, or iPhone 15 Pro Max from June 10, 2024, to March 29, 2025.
About 37 million iPhone owners are eligible to apply for the payment.
Each eligible device will receive $25, but this amount could change, potentially reaching up to $95.
Customers will be notified by email or mail if they qualify to file a claim.
A special website with claim instructions will be available in the coming weeks.
The lawsuit was filed in federal court in Northern California on behalf of U.S. consumers.
Apple stated the settlement helps them focus on creating new products and services.
Read the Original
Want the full story? Tap a source to open the original
article.
Shivon Zilis, an executive at Elon Musk’s company Neuralink and mother of four of Musk’s children, testified in a lawsuit Musk filed against OpenAI. Musk claims OpenAI’s leaders broke a founding agreement by changing the company from a non-profit to a for-profit, seeking damages and leadership changes. OpenAI denies these claims, and Zilis's role and relationship with Musk have become key points in the case.
Key Facts
Shivon Zilis worked at OpenAI from 2016 to 2023 and served on its board from 2020 to 2023.
Elon Musk alleges OpenAI’s CEO Sam Altman and president Greg Brockman broke an agreement when they restructured OpenAI into a for-profit company.
Musk wants $134 billion in damages and the removal of Altman and Brockman from OpenAI.
OpenAI says Musk agreed to the for-profit shift and suggests he left in 2018 after failing to gain control.
Zilis had a secret romantic relationship with Musk starting around 2016 and is mother to four of his children.
OpenAI’s lawyers say Zilis acted as an informant for Musk inside OpenAI after he left.
Text messages show Zilis coordinated with Musk about staying close to OpenAI and recruiting employees to Tesla.
Musk has referred to Zilis as his “chief of staff” and lives with her in Austin, where their children also live.
Read the Original
Want the full story? Tap a source to open the original
article.
Google’s DeepMind has taken a minority stake in Fenris Creations, the new independent developer of the popular sci-fi game EVE Online. They plan to use the game to test and improve artificial intelligence models in a complex, player-driven virtual world without affecting actual players. Fenris Creations recently bought itself out from its former owner Pearl Abyss and will continue developing EVE Online and new projects.
Key Facts
DeepMind has a minority ownership in Fenris Creations, the new name for the EVE Online developer.
EVE Online is a complex, player-driven game ideal for AI research involving long-term planning and learning.
DeepMind will run AI tests in a separate offline version of EVE Online to avoid disturbing real players.
Fenris Creations paid $120 million to become independent from South Korean publisher Pearl Abyss.
Pearl Abyss originally bought the EVE Online maker for $225 million six years ago, so the new sale reflects a lower company value.
Fenris Creations showed losses of about $20 million in 2023 and 2024 but became profitable in 2025 with $70 million in revenue.
The company plans to focus on new long-term projects, including blockchain and shooter spinoffs of EVE Online.
DeepMind has used games before to develop AI, including Go, StarCraft, and Atari games.
Read the Original
Want the full story? Tap a source to open the original
article.
NASA released over 12,000 photos taken by astronauts on the Artemis II mission to the Moon. An expert combined a series of "Hello, world" images into an animated sequence showing details like lightning storms, auroras, and satellites near Earth.
Key Facts
Artemis II astronauts took thousands of photos using Nikon D5, Nikon Z9, and iPhone 17s cameras.
NASA shared all these photos recently on the Gateway to Astronaut Photography website.
The "Hello, world" image was taken by Commander Reid Wiseman as the spacecraft left Earth.
Photographer Andy Saunders created an animation from 17 consecutive photos of this sequence.
The animation shows weather phenomena like lightning and auroras as well as satellites orbiting Earth.
Some visible satellite parts may be optical effects caused by the spacecraft’s window.
The animation speed is 30 times faster than real time, covering about 1 minute and 20 seconds.
Read the Original
Want the full story? Tap a source to open the original
article.
Anthropic has introduced a new feature called "dreaming" to its Claude Managed Agents. This feature reviews past events to find important information to save in memory, helping the agents work better on long projects and tasks.
Key Facts
"Dreaming" helps Claude Managed Agents review and save important memories from past sessions.
It is designed to improve multi-agent teamwork over long periods, like several minutes or hours.
The feature is in a research preview and not yet available to all developers.
Dreaming finds patterns such as repeated mistakes and common workflows across different agents.
Users can choose to let dreaming run automatically or review memory updates themselves.
Anthropic also expanded access to other features like outcomes tracking and multi-agent orchestration.
The company increased usage limits for its Pro and Max subscription users to address demand issues.
Managed Agents are a higher-level tool that runs in controlled infrastructure, making it easier to handle complex tasks.
Read the Original
Want the full story? Tap a source to open the original
article.
Many people use artificial intelligence (AI) to help with fitness, but opinions on it vary. Some find AI helpful for creating workout plans and keeping routines, while others worry about its accuracy and broader effects on society.
Key Facts
Around 62% of people in the US and 69% in the UK use AI regularly.
Only 26% of Americans have a positive view of AI, and 78% of people in the UK worry about its negative impacts.
Experts advise checking AI workout advice with real professionals due to possible errors that could cause injuries.
Some users share positive experiences, like sticking to routines and personalized plans.
Others criticize AI for being wrong or less helpful than human trainers.
AI can deliver information quickly, but users should use it carefully.
Personal stories show a range of attitudes toward AI fitness tools, from reliance to distrust.
Read the Original
Want the full story? Tap a source to open the original
article.
Google has improved the speed of its local AI models called Gemma 4 by using a new technique called Multi-Token Prediction (MTP). This method guesses several future words at once, allowing the AI to respond up to three times faster on various devices while keeping the same level of accuracy.
Key Facts
Google released the Gemma 4 AI models for local use, allowing users to run AI on their own devices.
Gemma 4 models use the same technology as Google's larger Gemini AI but are optimized for personal hardware.
Multi-Token Prediction (MTP) helps the AI generate multiple tokens (words or pieces of words) ahead of time to speed up responses.
MTP uses smaller, faster "drafter" models to guess future tokens, which are then checked by the main, larger model.
This process reduces the waiting time by producing tokens in parallel rather than one by one.
The speed increase varies by device, with some phones seeing nearly a 3x boost and Apple’s M4 processor seeing 2.5x faster performance.
Google changed the license for Gemma 4 to Apache 2.0, making it easier for developers to use and modify.
MTP helps overcome hardware limits in memory speed by reducing the need to move data repeatedly during AI processing.
Read the Original
Want the full story? Tap a source to open the original
article.
Apple will pay $250 million to settle a lawsuit that said it misled customers by advertising artificial intelligence (AI) features in iPhones that were not yet ready. This settlement could mean some iPhone users will receive payments.
Key Facts
Apple agreed to pay $250 million to settle the lawsuit.
The lawsuit was a class-action case filed in a federal court in San Francisco.
The case claimed Apple advertised AI features that were not yet available on iPhones.
The settlement applies to U.S. consumers.
Some iPhone users may get payouts as part of the settlement.
The issue involved the AI assistant called Siri.
The lawsuit focused on claims of deceptive advertising by Apple.
Read the Original
Want the full story? Tap a source to open the original
article.
NASA wants to land on the Moon up to 21 times in the next two and a half years. To do this, NASA needs better management of moon lander projects, improved industrial oversight, and more reliable deliveries from its commercial partners.
Key Facts
NASA plans about 21 Moon landings between now and 2027, including crewed and robotic missions.
The Commercial Lunar Payload Services (CLPS) program contracts private U.S. companies to build lunar landers.
Since 2019, NASA has contracted 13 lander missions, but only one has landed successfully so far.
Commercial lunar lander projects often face delays, with some missions possibly slipping into 2027.
NASA wants to improve the reliability of landers by closely working with and supporting its industrial partners.
The agency plans nine lunar landings in 2025 and ten in 2026 to build a Moon base and test new technologies.
NASA aims to use lessons from its previous commercial space programs to speed up lunar lander development.
NASA Administrator Jared Isaacman emphasizes moving beyond planning to faster, better results with commercial teams.
Read the Original
Want the full story? Tap a source to open the original
article.
Apple has agreed to pay about $250 million to settle a U.S. class action lawsuit claiming the company misled customers about AI features in new iPhone models. The case focused on Apple’s marketing of advanced Siri capabilities that were not fully available when the iPhones shipped.
Key Facts
The lawsuit was filed in March 2025 against Apple for false advertising about AI features in iPhone 16 and some iPhone 15 Pro models.
Apple promoted upgraded Siri features as part of a new “Apple Intelligence” system, but many promised functions were delayed until 2026.
The settlement covers U.S. buyers of affected iPhones purchased between June 10, 2024, and March 29, 2025.
Eligible consumers may receive between about $25 and $95 per device, depending on the number of claims.
Millions of devices (around 36 million) might qualify, but no filing deadline has been set yet.
Claimants need proof of purchase and device details to apply for compensation after judge approval.
The U.S. Federal Trade Commission warned companies in 2024 that AI features must not be advertised misleadingly, emphasizing consumer protection laws apply to AI marketing.
This settlement may create a legal precedent about truthful advertising of AI features in technology products.
Read the Original
Want the full story? Tap a source to open the original
article.
CBS News' Anthony Mason is starting a new YouTube show called "Alchemy with Anthony Mason." The show will feature long interviews with artists and explore how they create their work.
Key Facts
Anthony Mason is a journalist working with CBS News.
He is launching a YouTube show named "Alchemy with Anthony Mason."
The show will have long interviews, meaning they will go in-depth.
It will focus on artists and how they create their art.
The program aims to celebrate creativity and the creative process.
The show will be available on YouTube, a popular video platform.
Viewers can watch it through various browsers like Chrome and Safari or the CBS News app.
Read the Original
Want the full story? Tap a source to open the original
article.
The new video game 007 First Light, developed by IO Interactive, offers a fresh take on the James Bond character, focusing on his early career. The game mixes stealth and action gameplay similar to Hitman, with major fight and chase scenes, aiming to appeal to modern audiences. Critics noted some guided gameplay and a talkative Bond actor, but the game is seen as a potential revival of Bond in the gaming world.
Key Facts
007 First Light is developed by IO Interactive, known for the Hitman series.
The game follows a young James Bond during his first mission as a double-0 agent.
Gameplay includes stealth sections, large shootouts, chase scenes, and gadgets.
Some players found the action too guided, similar to other story-driven games like Uncharted.
Patrick Gibson voices Bond but some reviewers found his portrayal overly eager and chatty.
Previous Bond games struggled to capture the full style and charisma of the movies.
007 First Light aims to give players freedom in how to play Bond, including using charm or fighting.
Amazon owns the Bond franchise and chose this game as its first major new Bond project.
Read the Original
Want the full story? Tap a source to open the original
article.
Iran used online videos with Lego-like and cartoon characters to mock President Donald Trump's decision to pause a U.S. operation in the Strait of Hormuz. These videos are part of an information campaign that uses artificial intelligence (AI) to create viral propaganda.
Key Facts
President Trump paused a U.S. operation in the Strait of Hormuz.
Iran's online groups quickly claimed this as a win and mocked the President.
The propaganda includes videos made with Lego-style, Minecraft, Minions, and Simpsons-like characters.
These videos are created using AI technology to produce viral content.
The videos aim to influence public opinion, especially in the U.S.
This is part of a broader use of AI-generated fake videos in the Iran conflict.
The effort shows how modern conflicts involve digital information tools beyond traditional warfare.
Read the Original
Want the full story? Tap a source to open the original
article.
The article discusses the author’s experience with using command-line interfaces (terminal windows) over the years. The author explains that despite the popularity of graphical interfaces, command-line tools remain very useful and precise, leading the author to switch from Windows to MacOS because it offered both a graphical interface and a strong command-line environment.
Key Facts
The author spent a lot of time using terminal windows and command-line tools in daily work.
Early graphical interfaces like Windows seemed to replace command lines but text commands remain important for many tasks.
The author started using the bash shell at work to manage enterprise storage systems.
Windows Vista had technical issues that pushed the author to consider other operating systems.
MacOS was chosen because it combined a good graphical interface with a powerful Unix-style shell.
The author now uses MacOS as the main desktop system, Windows for gaming, and Linux as a headless server.
The fish shell (a modern command-line shell) is the author’s current favorite on MacOS due to its helpful colors.
The article includes before-and-after photos of the author’s terminal setup showing customization.
Read the Original
Want the full story? Tap a source to open the original
article.
The article says that after the Artemis II mission, the next goal for human spaceflight should focus on exploring Mars. It suggests that going to Mars will be the best way to advance space travel for people.
Key Facts
Artemis II is a space mission focused on human spaceflight.
The article recommends using the success of Artemis II as a step toward Mars missions.
Mars is called the red planet because of its color.
The goal is to expand human space travel beyond the Moon.
Exploring Mars could help improve space programs and technology.
The article emphasizes Mars as the future target for human missions.
Human spaceflight means sending astronauts to space destinations.
Read the Original
Want the full story? Tap a source to open the original
article.
Rivian is preparing to launch its smaller, more affordable R2 electric SUV, competing with models like the BMW iX3. The company plans to build the R2 and a smaller R3 model at a new factory in Georgia, which is expected to start production in 2028 with a large capacity to make many versions of these vehicles.
Key Facts
Rivian has released larger electric vehicles like the R1T pickup and R1S SUV, and makes Amazon delivery vans.
The R2 is a midsize electric SUV that will cost less and compete with cars like the BMW iX3.
A $45,000 version of the R2 with a smaller battery may arrive next year.
Rivian CEO RJ Scaringe hinted at more versions of the R2, such as an R2X.
Rivian is thinking about making its own lidar sensors, which help cars see their surroundings, in partnership with a Chinese firm.
Initial R2 models will be made at Rivian’s Illinois factory, which can produce about 155,000 vehicles a year.
A new factory in Georgia is planned to open in 2028, with a capacity of 300,000 vehicles annually, supported by a $4.5 billion loan from the U.S. Department of Energy.
The Georgia plant will also build robotaxis for Uber under a $1.25 billion deal.
Read the Original
Want the full story? Tap a source to open the original
article.
Scale AI CEO Jason Droege says that AI systems often make mistakes that are too costly for important uses in business, military, and government. He focuses on making AI more reliable by combining technology with human expertise to meet customer needs.
Key Facts
Scale AI recently celebrated its 10th anniversary and is based in San Francisco.
Jason Droege became CEO in June last year, after founder Alexandr Wang left to work at Meta.
Meta now owns 49% of Scale AI.
Droege wrote a memo called "The Reliability Race" to stress the importance of AI reliability for customers.
Scale employs over 1,300 people, including engineers who work closely with clients to ensure AI works well in specific use cases.
Customers want AI that consistently performs well without large mistakes that cause distrust.
Scale AI aims to bridge the gap between the hype around AI and its real-world performance.
Droege focuses on clear priorities to lead the company through a rapidly changing industry.
Read the Original
Want the full story? Tap a source to open the original
article.
