Wired: Sources: Meta internally exposed data from its employee-tracking program meant to help train its AI models, including full prompts and private conversations  —  Employees had previously raised concerns about the initiative, which involves collecting workers' keystroke data to train AI models.

He had retinal tears and bruises from squishing his eyeballs with the gun.

The loop takes agentic AI a step further by authorizing a swarm of agents to work continuously in the background, endlessly.

Tesla has faced scrutiny over its driver-assistance technology. John Keeble/Getty Images A Tesla Model 3 crashed into a home in Katy, Texas, killing a resident inside on Friday night. Driver Michael Butler said Tesla's driver assistance tool was active during the fatal crash. A Tesla executive said the driver 'manually overrode' the self-driving feature by speeding. Federal regulators have opened a probe into a collision that killed a Texas woman when a Tesla crashed through a brick home last week. The driver reported an automated driving-assistance system was engaged at the time. A Tesla official said the driver "manually overrode" the self-driving feature with heavy acceleration. A spokesperson for the National Highway Transportation Safety Administration confirmed that a "Special Crash investigation" — the "most in-depth and detailed level of crash investigation data" collection — has been launched. The federal regulator said that the cases NHTSA chooses could involve probes into "emerging technologies," including "alternative fueled vehicles," "adaptive controls," and "potential safety defects." The crash happened around 8:03 p.m. on Friday in Katy, Texas, where Michael Butler was driving a Tesla Model 3, the Harris County Sheriff's Office said in a statement on Saturday. Authorities said the driver told them he had an automated driving-assistance system engaged. "Butler failed to drive in a single lane, left the roadway, and struck the residence," the sheriff's office said. "Butler's Tesla entered through the brick residence, at a high rate of speed, and struck M. Avila who was inside the residence," it added. Avila was flown by medical helicopter to a local hospital, where she later died from injuries sustained in the crash, according to the sheriff's office. "There were no signs of intoxication on Butler and he was cooperative during the investigation," the statement added. In an X post on Monday, Ashok Elluswamy, Tesla's head of AI, said that the driver "manually overrode self-driving by pressing the accelerator all the way to 100% of the accel pedal in this residential area." "They reached a speed of 73 mph during the crash, and had the accelerator pressed even after the crash," he said in the post. The Harris County Sheriff's Office did not immediately respond to an inquiry regarding Elluswamy's post. It was not clear which of Tesla's features Butler was using at the time of the crash — the company offers Autopilot and full self-driving capabilities. Tesla says the features are for the "fully attentive driver" who should be ready to take over at any time. Tesla did not respond to a Business Insider request for comment. Business Insider was unable to contact Butler. The crash comes as Tesla's driver-assistance features have faced scrutiny in courts and from regulators. Last year, a Florida jury found the automaker partially liable in a 2019 crash in which a driver hit another vehicle while Autopilot was engaged. In a separate case, a California judge ruled last year that Tesla's branding of the feature as "Autopilot" was misleading, leading the company to change its name. Update: June 22, 2026 — This story has been updated to reflect that the driver told authorities the automated driving-assistance feature was engaged at the time of the crash; the investigation is ongoing. It also adds a statement from a Tesla executive about the circumstances of the crash. Read the original article on Business Insider

Amazon CEO Andy Jassy Brendan McDermid/Reuters Amazon buys ads on ChatGPT to drive users back to its own marketplace. Amazon blocks AI companies from scraping its product and pricing data. OpenAI's ad business could get a major boost from Amazon's presence. Amazon wants ChatGPT's users. It just doesn't want OpenAI to have its data. Amazon has begun buying ads on ChatGPT, making the company one of the highest-profile retailers to appear in OpenAI's emerging advertising business, according to e-commerce analyst Juozas Kaziukėnas. The ads direct users back to Amazon's storefront, where the company controls the customer experience and transaction. Kaziukėnas told Business Insider that Amazon's decision to buy ads on ChatGPT is "symbolic" because the retailer has largely avoided participating in AI shopping initiatives that let third-party chatbots and AI agents aggregate its products, pricing, and inventory data. The move highlights a tension at the heart of Amazon's AI strategy. The company is willing to pay to reach ChatGPT's massive user base, but it continues to protect its own shopping data and limit how AI systems can access and use it. Rather than helping AI platforms aggregate Amazon's catalog, the e-commerce giant is using them as marketing channels that send customers back to Amazon's online marketplace. An Amazon spokesperson declined to comment. Kaziukėnas initially shared a screenshot of Amazon's ad on ChatGPT through a Linkedin post on Monday. It showed the chatbot suggesting several products under a search for a coffee maker, followed by a sponsored Amazon ad promoting "Top-Rated Kitchen Gear." Amazon ad on ChatGPT Juozas Kaziukėnas Amazon has taken steps to restrict AI scraping and data collection efforts that could be used to build competing shopping experiences. Last year, Amazon stopped providing product feeds to Google Shopping results while updating its code base to block multiple bots, including those from OpenAI, according to Kaziukėnas. Earlier this year, Amazon won a court order to block Perplexity's AI agent. Amazon's arrival on ChatGPT could also be an encouraging sign for OpenAI's nascent advertising business. Early ad data has shown that ChatGPT users frequently encounter ads when making commercial-intent queries, and advertisers are increasingly viewing the chatbot as a new channel for reaching consumers who are actively researching products. "To me, this is a sign that OpenAI will have an easier time monetizing shopping intent through ads than the still unproven agentic commerce, which it abandoned," said Kaziukėnas. "Its ads business will grow fast." Have a tip? Contact this reporter via email at ekim@businessinsider.com or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely. Read the original article on Business Insider

Microsoft inked a 20-year power purchase agreement with Chevron, locking in decades of carbon emissions from a new natural gas power plant.

Employees had previously raised concerns about the initiative, which involves collecting workers’ keystroke data to train AI models.

iPhone 18 Pro rumors point to AI upgrades, a 2nm A20 Pro chip, camera changes, a smaller Dynamic Island, and possible pricing shifts. The post Apple’s iPhone 18 Pro Features: Everything We Know So Far appeared first on TechRepublic.

The Army activated the Multi-Domain Command Pacific last week, combining the 7th Infantry Division and 1st Multi-Domain Task Force. US Army video by Staff Sgt. Brandon Rickert The Army merged the 7th Infantry Division and 1st Multi-Domain Task Force into a new Pacific-focused command. The command combines maneuver forces with drones, sensors, cyber, and long-range fires. The Army says the unit is built to operate in the Pacific as a self-contained force. The US Army has merged two formations into a new Pacific-focused command built to move fast, operate with limited support, and effectively bring drones, electronic warfare, and long-range fires into a future fight. The command, shaped by years of experimentation, including drills with regional allies, is meant to give the Army a more self-contained force for complex operations across the Pacific. Army officials announced the establishment of the service's new Multi-Domain Command-Pacific last week, a one-of-a-kind command combining the service's 7th Infantry Division and 1st Multi-Domain Task Force into a new force that brings together traditional maneuver capabilities and weapons across domains like cyber, electronic warfare, space, and long-range fires. The command, headquartered at Joint Base Lewis-McChord in Washington, is composed of around 12,000 soldiers, 7th ID's Stryker brigades, which are mechanized infantry forces, and the Multi-Domain Task Force's capabilities. Army officials said the merger would enhance the Army's ability to fight in denied or degraded environments. Maj. Gen. Bernard Harrington, who leads the new command, shared in a release that the merger brings together the "operational endurance, flexibility, and protection" of Stryker formations with the "long-range sensing and precision fires" of the Multi-Domain Task Force. "Through our emerging cross domain contact layer concept, our division will employ capabilities such as unmanned surface vessels; long-range, one-way attack drones; and launched effects to penetrate the adversary's anti-access/area-denial network," he said. The new command "allows us to see, sense, make sense, affect, protect, and sustain a force forward in the Pacific region," Gen. Ronald P. Clark, commander of US Army Pacific, told reporters last week during a media roundtable discussion. Under this framework, conventional units in the 7th ID would be enabled by the multi-domain weapons of the task force, and in turn could protect and move those weapons, officials said. The Army tested the concept at Balikatan 2026, the US' large-scale annual exercise with the Philippines, before rolling it out officially. US Army photo by Staff Sgt. Yesenia Carrero The Army put the concept to the test during Balikatan 2026, a large-scale exercise led by the US and the Philippines, and then officially activated the command last week. During Balikatan, which ended last month, "we used Army watercraft for sustainment, we utilized our unmanned systems," Harrington shared with reporters. "We networked them all together and really provided those multi-domain effects in a real theater environment as a rehearsal of concept." The Army's new Multi-Domain Command-Pacific also draws on lessons from the Ukraine war and other conflicts, where drones have been used for long-range intelligence, surveillance, and reconnaissance (ISR) and strikes. Officials said that they wanted forces to be able to employ a large mass of drones that could overwhelm enemy systems. The Army and US military more broadly have been working on integrating a variety of drones into Pacific warfighting. The integrated task force "is sensing at those very long ranges," Harrington said, explaining that "it is passing that data to shooters to be able to engage at those very long ranges, and it is all doing that on behalf of the theater Army." Essential to that effort is the Army's cross domain contact layer, which connects ISR capabilities, artificial intelligence, and electronic warfare into a continuous sensor network that, particularly in the vast Pacific region, can help the Army spot potential threats at greater distances and determine the best way to respond. The Multi-Domain Command-Pacific change comes amid broader transformation efforts across the Army focused on new technologies and weapons, as well as integrating AI to prepare forces for wars where speed will be a priority, soldiers may be overwhelmed with data from sensors, and critical systems could be consistently denied or degraded. Read the original article on Business Insider

X product head Nikita Bier targeted “neglected” Meta employees with a snack-budget hiring pitch as Meta works to improve morale. The post X Targets ‘Neglected’ Meta Engineers in Public Recruiting Pitch appeared first on TechRepublic.

Meta founder, chairman, and CEO Mark Zuckerberg. Tom Williams/CQ-Roll Call, Inc via Getty Images Data from Meta's AI training program was accessible across the company, screenshots showed. The leak, which exposed employee conversations and keystrokes, is causing backlash within Meta. Meta said it has paused the training program while investigating the incident. Meta is pausing an internal AI training program after sensitive data was accessible across the entire company, according to screenshots obtained by Business Insider. A screenshot showed that the leak exposed employees' private conversations, performance data, and transcriptions. The incident was classified as a SEV 2 on a scale of 0 to 5, with 0 being the most severe. A Meta spokesperson confirmed the incident and said the company is investigating. "We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we're pausing it while we investigate," the spokesperson said. In April, Meta announced the AI training program, called the Model Capability Initiative (MCI), which was intended to improve the company's AI models by using its staff's keystrokes and mouse movements as training data. The program, which is mandatory for most staff, sparked a backlash from employees who felt uncomfortable with their data being recorded, Business Insider previously reported. This leak is causing frustration within Meta, according to screenshots seen by Business Insider, with employees critical that data wasn't locked down from the start. "I am incensed," one employee wrote on Monday about the recent leak in an internal group, according to a screenshot obtained by Business Insider. "I don't see any evidence of malicious access, but the fact that this data wasn't locked down as originally promised is super frustrating," the employee continued. The leak is the latest in a string of recent security incidents for Meta. Last month, a flaw in its AI chatbot allowed people to hijack multiple Instagram accounts. A rogue AI agent also caused a severe incident in March, the Information reported. Correction: June 22, 2026 — An earlier version of this story misstated Meta's severity scale. 0 is the most severe on its scale. Have a tip? Contact Charles via email at crollet@businessinsider.com or on Signal and WhatsApp at 628-282-2811. Contact Pranav via encrypted messaging app Signal at +1-408-905-9124, or email him at pdixit@businessinsider.com or pranavdixit@protonmail.com. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely. Read the original article on Business Insider

Allyson Versprille / Bloomberg: Air Space Intelligence won an $875M, 12-year FAA contract to develop AI tools that map flight trajectories and identify areas of congestion to reduce delays  —  Air Space Intelligence Inc. won a US government contract to develop artificial intelligence technologies for managing flight traffic …

Discussion | Link

What does an AI company do after one of those not-acqui-hire deals? Groq raised money, is leaning into its neocloud business, and is hiring new execs.

"Winning" bets were made on cloned website and would have lost money, WSJ finds.

Nvidia announced a new cooling system that cuts water use inside the data center. But it does nothing to address AI's biggest water use — fossil fuel power plants.

Ina Garten's tomato feta pasta salad is officially my new go-to summer side dish. Lizzy Briskin On the hunt for a new summer side dish, I tried Ina Garten's tomato feta pasta salad recipe. The dish was easy to make, and I found it more flavorful than other pasta salads I've tried. I doubled the recipe, and my household enjoyed leftovers for several days. As a chef, I'm a longtime fan of Ina Garten's classic, riffable recipes. They're usually easy to recreate at home, and often a huge hit with guests. I've also been looking for a simple side dish to bring to summer cookouts and picnics. When I found Ina Garten's tomato feta pasta salad, I knew I had to give it a shot. Here's what happened when I tested the dish out for myself. Spoiler alert: I'll be adding this one to my recipe rotation. The ingredients make this dish stand out from other mayo-based pasta salads. caption TK Lizzy Briskin The recipe calls for short-cut pasta, black olives, feta cheese, and two kinds of tomatoes: fresh and sun-dried. It's key to use sun-dried tomatoes packed in oil, as they're more tender and flavorful than the dry variety. As for the pasta shape, Garten recommends fusilli, which has a corkscrew-like shape that soaks up extra dressing. After trying out the recipe, though, I'm confident any short-cut shape could work. The olives (Garten recommends Kalamata), tomato, and feta all bring major umami to this summertime dish. The salty, briny mix-ins give the salad a distinctly Mediterranean taste, which I found I preferred over the typical mayonnaise-dressed pasta salad. For the dressing, I grabbed more sun-dried tomatoes, red wine vinegar, olive oil, capers, and garlic. I'd also recommend some freshly grated Parmesan and parsley to top the salad off. The prep work was quick and easy. caption TK Lizzy Briskin As is the case with every good pasta-salad recipe (in my opinion, anyway), prepping the ingredients doesn't take a ton of time or effort. While I boiled a big pot of salted water for the pasta, I chopped the fresh tomatoes, sliced the olives, and diced the feta and sun-dried tomatoes. Garten calls for diced whole tomatoes, but I used cherry tomatoes and simply sliced them in half for the same effect. The dressing calls for a food processor, but it's possible to make it without one. caption TK Lizzy Briskin The best way to make this salad dressing is with a small food processor. You could also try an immersion blender or pitcher-style blender, but expect to spend some time scraping down the sides. By whizzing more sun-dried tomatoes, red wine vinegar, olive oil, garlic, and capers in a food processor, I was able to break down the capers and tomatoes to maximize their flavor output. I ended up with a fairly smooth, reddish dressing that was thin enough to coat every nook and cranny of the pasta. However, if you don't have any blade-loaded countertop appliances, you should be fine — just give everything a fine chop and a good stir. The dressing won't be as smooth and creamy as Garten intended, but your pasta will still benefit from the flavor-packed ingredients. I dressed the pasta while it was still hot from the stove. caption tk Lizzy Briskin My pasta finished cooking in the time it took to prepare the dressing. Although Garten recommends letting the noodles cool first, my pasta-salad-making experience has taught me that tossing hot noodles in an oily, salty dressing maximizes flavor. I tossed the cooked pasta, salty mix-ins, and dressing together in a large bowl. I then let the pasta cool to room temperature before finishing the dish with Garten's recommended chopped fresh parsley and a generous amount of grated Parmesan. Here, I made sure to add the cheese after the salad cooled so it wouldn't all melt into a gooey mess. Instead, the Parmesan coated the noodles, mimicking the creamy, rich dressing many of us associate with pasta salad. Compared to other salads, though, this one is big on savory flavors and pops of briny, chewy sun-dried tomatoes. I loved that the dish felt lighter than a mayo-drenched salad, but just as satisfying. The oil-based dressing holds up well, too. I doubled the recipe, and my household of two enjoyed pasta for several days. I'm glad I found this recipe right before cookout season — I'll be bringing this easy, crowd-pleasing dish to every picnic and summer party this year. This story was originally published on July 4, 2025, and most recently updated on June 22, 2026. Read the original article on Business Insider

Carl Franzen / VentureBeat: Sakana AI launches Fugu, a multi-agent orchestration system accessible through a single model API, claiming Fugu Ultra matches Fable and Mythos on benchmarks  —  Last night, the increasingly enterprise-focused AI startup Sakana launched Fugu, a multi-agent orchestration system …

Discussion | Link

MrBeast, whose real name is Jimmy Donaldson, has been on a tech hiring spree at his company, Beast Industries. Kristina Bumphrey/WWD via Getty Images MrBeast's company hired a team from the Andreessen Horowitz-backed startup Pietra. They'll help Beast Industries build a platform to match brands and creators. The company is racing to diversify itself beyond YouTube videos. MrBeast has been on a hiring spree as he builds a platform to match brands with creators for sponsorship deals — and diversify his company beyond YouTube. His company, Beast Industries, recently hired a team from the creator-commerce startup Pietra, including its CEO and cofounder Ronak Trivedi, Business Insider has learned. Trivedi, who worked at Uber before helping found Pietra, is playing a key role in developing the yet-to-be-named creator platform, a Beast Industries spokesperson confirmed. The rep didn't share the number of hires but called it "significant." Pietra has raised $36 million to date from Andreessen Horowitz and others, according to Crunchbase. It helps creators develop their own consumer product lines by connecting them with product designers, suppliers, warehouses, and fulfillment companies. The MrBeast spokesperson said Pietra would continue to operate under a new CEO. Elsewhere, Beast Industries recently hired a chief product and tech officer, Google and Uber vet Shiva Rajaraman for the company. Beast Industries said it has also opened a product and engineering office in San Mateo, California, to work on the creator platform and a planned membership program. MrBeast, whose real name is Jimmy Donaldson, has been expanding beyond his YouTube channels for a while, into merchandise, a snack line called Feastables, financial services, and more. Donaldson recently announced plans for a membership program and a mobile phone venture. The company has hinted at a potential IPO. Jeffrey Housenbold, CEO of Beast Industries, spoke about the creator platform onstage at The New York Times' DealBook Summit in December, describing it as a way to connect creators with Fortune 1,000 marketers. A February 2025 investor deck said it was a way to help creators replicate Donaldson's success by giving them "data-driven opportunities" to grow their channels and viewership. As YouTube's biggest creator, Donaldson offers credibility in matching creators and marketers. He already serves brands and creators with the YouTube analytics platform Viewstats and Vyro, a clipping service that gives brands access to a large creator network. Marketers have been shifting budgets from traditional media to social media creators. The Interactive Advertising Bureau estimated in November that advertisers would spend $43.9 billion on creators in the US in 2026, almost doubling their 2023 spend. Donaldson would be entering a crowded field, as holding companies, tech platforms, and independent agencies already offer similar services to varying extents. The Pietra hirings come at a time of change for Beast Industries' tech staff. Several people have recently left Viewstats and Vyro, including Nagesh Nagpal, who was COO at Viewstats. The company spokesperson said those departures were unconnected to the planned creator platform and that Viewstats and Vyro would continue to operate as separate units. Read the original article on Business Insider