• Jay Eskenazi planned the itinerary for his trip to Europe with Claude and the Grok Bot.
• Once he landed, Eskenazi said he realized Meta's Muse was the most helpful for travel planning.
• "These are the things that a very seasoned travel agent from 30 years ago would've done for you," he said.

This as-told-to essay is based on a conversation with Jay Eskenazi. He lives in Los Angeles and worked in user experience for 19 years. It's been edited for length and clarity.

I'm in Paris, and I'm going to Barcelona tomorrow. The trip is three weeks.

I came here for the first time last year, and it was my first long trip by myself. Last year, planning it was very stressful, and I did a ton of research.

This year was the same thing, where I had these very long notes. At this point, everyone knows that you can go to any AI tool and say: "Hey, I'm going to City X, give me an itinerary for seven days."

I had this epiphany a few weeks ago, where I realized I had all these note files, bulleted stuff, and coffee shops to go to.

If someone gave you a list of five coffee shops and they were in five different cities, as someone who's not familiar with the terrain, you wouldn't necessarily know how to structure that in the course of the day. Which coffee shop goes with which museum?

So, instead of doing an itinerary, I was like: "Here's my notes. Put them together in an organized way for me. Structure them in a logical sequence."

I hadn't used Claude in a while, so I tried that a little. I use Grok a lot, and I just upgraded because I wanted to play with the Grok Bot. So, I made this really fancy itinerary. I kept iterating it. It did a really good job. Then I got here and was like, "It's not as good."

I'm naturally curious, so I tried out Muse. I think I must've gotten Muse two days before I left.

I was looking at this itinerary a couple of days ago and was like, "I don't like what it says tomorrow." So I opened Muse and asked for an afternoon itinerary. It spat out all of this stuff. I was like: "Oh! There's some new stuff in here."

These are really common scenarios. Maybe there are five museums you want to go to. It takes a lot of work: this museum's closed on Tuesday, this one's closed on Monday, and this one's free next Friday. The AI tools can structure that in a way that you would do if you had a thousand hours.

It's a small thing, but when you apply that to a thousand decision points, it does a really nice job of putting those things together.

If you're like: "Hey, I want to go get coffee, then I need something to do before I get lunch, then I want to go to the museum this afternoon." Most of the other AI tools will just give you the list. Then you're like, how do I get from place to place to place?

Muse would actually put it on a map with dots. I clicked a button, and it put it in Google Maps. I had this experience this weekend. None of the other tools did that.

I'm going to Barcelona tomorrow, and I'll get there in the evening. One of the reasons I picked it was because there's an annual festival. I timed it so I can be there for the core of the activities. But the main activities weren't published, even last week, because they're on some website in Spanish that didn't have the fully fleshed-out program.

In the hourlong Uber to Versailles, I was like, "What's the structure for Thursday?" It's able to pull that. It's got: this opens in the morning, this is free, this is mid-day, this is in the evening.

I don't know if ChatGPT would have it. Maybe it would, but it doesn't seem to output it in the same way. Also, I gave it access to my email, so it knows I'm leaving tomorrow. It has an idea tab that said, "I can map the opening night before you get there." It's proactive.

The other thing: I was indecisive about my plans over the last couple of days. What Muse was doing was checking the times to see what ticket availability looked like. It's doing this proactive thing for you, where it tracks it, because options are shrinking.

As a non-developer, I set up OpenClaw in February. That took a lot of time. I had a developer friend help me for hours. Then I had all these AWS bills I didn't understand. I'm asking it to give me a recipe; I don't know how many tokens that's going to cost. I ended up having to shut it down.

Muse is like OpenClaw for normies. If you're not a tech person, these things are useful because they do what an assistant would. Anyone could set this up.

It comes back to when I worked at Expedia. These are the things that a very seasoned travel agent from 30 years ago would've done for you. That would be a very expensive vacation for that level of service.

I also just got access to Instinct yesterday, because I was on the waitlist. I'm very curious, but you can't see exactly what it's doing. It had good ideas and gave me an itinerary, but it had to put them in a Google Doc. I was also confused because I don't remember what I told it, and it made personalized recommendations. I was like: "Wait, how did it know this?"

Muse just seems easier. It's very fast, and it feels more responsive.

• OpenAI boosts ChatGPT's voice mode on mobile as the competition over personal agents heats up.
• The new release connects OpenAI's voice model to its agentic text models, which can handle tasks.
• OpenAI has yet to release a direct competitor to viral AI assistants like Instinct and Meta's Muse.

OpenAI is pushing further into the hotly contested market for personal agents — this time with voice.

The company is adding its most advanced voice system to ChatGPT's website and mobile app on Wednesday. OpenAI's release allows users to speak to delegate single and recurring tasks that require multiple steps, a browser, and disparate apps. The company has been rolling out launches ahead of its big OpenAI DevDay event next week.

OpenAI's product lead for ChatGPT voice, Atty Eleti, gave a demo to Business Insider. Only speaking with the app, he had the tool analyze his DoorDash spending, pay $5 to book a pickleball court, and reorganize his calendar.

"In the future, we really imagine people using voice as the primary modality to interact with AI," Eleti said, "and that's kind of our North Star that we're working towards."

Personal AI assistants like Instinct and Meta's Muse are having a moment in Silicon Valley as the industry peddles apps tailored to handle cumbersome internet tasks. OpenAI has many of these capabilities in ChatGPT, though it hasn't yet released a narrower offshoot. This voice tool is a play for the AI-assistant buzz.

In the short term, the improvements make it easier for people to delegate work to ChatGPT by speaking while they're on the go. In the long run, they'll serve OpenAI's hardware ambitions: Bloomberg has reported that the company is building a smart speaker for release in 2027.

Wednesday's release connects OpenAI's voice model, GPT-Live, to AI models that can perform tasks on computers or in browsers, a focus the company maintained before the personal AI agent craze. GPT-Live, which can listen and respond simultaneously, was already turned on for ChatGPT's basic conversational mode and connected to agentic models on the desktop app.

Voice mode, as Eleti sees it, has three main benefits. Speaking is quick and natural — "you can blurt out your thoughts," he said. You can do it with your hands off a device. And it can help you with vocal tasks, like learning a language or practicing a speech. Eleti and his colleagues use it while commuting and on walks, he said.

Eleti said more than 150 million people already use ChatGPT's dictation and voice mode tools. Still, the team faces an uphill battle. Though smart speakers have shown some staying power, offices are still organized around desktop computers.

Plus, though it's quicker to speak than type, it's also quicker to read than to listen. Eleti said his team considered this dynamic when deciding how ChatGPT would respond to audio prompts.

"We don't think of this as just pure voice in and voice out," Eteli said. "It's about voice in and then the right output back. Sometimes it's text, sometimes it's voice, sometimes it's a mix."

Have a tip? Contact this reporter via email at scouncil@businessinsider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• A growing share of Americans living in rural areas aren't happy with data centers.
• Most data centers are built in rural areas.
• Data centers are facing a public reckoning amid bipartisan pushback.

Americans living in rural areas are increasingly souring on data centers.

In a new poll by the Pew Research Center, which surveyed people in August, half of rural Americans said data centers "are mostly bad for the environment" — a figure substantially higher than in January, when it sat at 32%.

46% of those respondents also said that data centers would drive up home energy costs, up from 32% in January. 45% said they negatively impact the quality of life for those living nearby, while 28% said so in January.

While displeasure with data centers increased among Americans across the board, according to Pew, higher levels of antagonism among rural voters are striking because most data centers are built in rural areas.

The data center buildout, which proponents say is integral for the US to keep pace with China in the AI race, has been suffering a PR crisis as a bipartisan coalition of lawmakers moves to impose limits on data center construction.

Texas Gov. Greg Abbot, who was formerly bullish about building data centers in his state, recently backed away from that optimism. His administration has halted some 1,800 AI data center projects in the state.

All across the country, Americans are rallying to fight against AI data centers, which can cost billions to build and span hundreds of acres. Many are concerned about how the sprawling facilities will affect the environment, utility prices, water resources, and noise levels. Some just don't like AI.

For AI companies like OpenAI, Anthropic, Meta, Google, and others, the issue is existential. Without AI data centers, they can't power the products they sell. Sam Altman has recently suggested that new data centers could be built far from where people lived, such as the desert.

Proponents of data centers have said the buildout brings jobs to rural areas. Venture capitalist Gavin Baker recently said that data centers are "probably the best thing that has ever happened to working-class Americans."

Tech bros haven't been able to agree on exactly why Americans aren't taking to the data center buildout. Some have postulated that a Chinese-orchestrated psy-op is behind the distrust, while others said tech leaders haven't sold the product well.

President Donald Trump, for his part, called the data center buildout "the golden goose," and warned against falling behind China.

"The only reason that communities throughout the U.S.A. should not want Data Centers is if they want to end up being backwards and poor," Trump said in August in a Truth Social post.

In the Pew poll, very few Americans — about 4% — said data centers are "good" for the environment, energy costs, and quality of life.

"Americans have grown more negative toward the possible environmental and economic effects of data centers since the beginning of the year," Pew said in a release. "And a majority say they'd be uncomfortable if a new data center started operating in their area."

• Sumitomo CEO Shingo Ueno believes 'kosoryoku' is a key skill in business.
• The term translates as 'the ability to envision futures that do not yet exist'.
• Ueno says AI can be a helpful research tool, but cannot yet conceptualize the future well.

A little-known Japanese term about imagining the future helps explain both the role — and the limits — of AI in the workplace, according to the CEO of one of the country's largest conglomerates.

Speaking to Business Insider in early September, Shingo Ueno, the head of Sumitomo Corporation, a Japanese multi-industry powerhouse with a market capitalization of around $55 billion, said that AI is a powerful research tool but lacks a crucial human capability: kosoryoku.

Pronounced koh-soh-ryoh-koo, the Japanese term essentially means the power to conceptualize or formulate a vision for the future that has not yet been imagined, something Ueno says AI does not currently have.

"Generative AI is helping me a lot to improve my efficiency. I'm fully utilizing it," Ueno said.

One of its biggest advantages is its ability to retrieve information quickly. "If I want to find information about something the company did before, I might have previously spent a day searching for that experience. Now, the answer comes immediately," he said.

That speed helps Ueno, who has spent over 40 years working for Sumitomo Corporation, follow his preferred management style: resolving issues quickly instead of allowing them to linger.

"My method is to make decisions on the spot," he said. "I try to avoid the back-and-forth — having the same issue come back the next day."

Still, Ueno has found that AI cannot make every decision for him. "I once tried to make a decision using AI, and I just got rubbish answers," he said. "It wasn't usable at all."

The problem, he said, is that AI depends on information from the past. It can identify patterns and suggest possible directions, but it is less capable when it comes to envisioning a future that does not yet exist.

"AI analyzes past data to show us a direction. But it is still based on the past," Ueno said. "It can't create the future. It can't really show the vision."

Developing kosoryoku does not require employees to come up with revolutionary ideas from scratch, he said.

"We don't need to be inventors. We don't need to come up with something completely new," Ueno said. "We can start with something that already exists, improve it, and make it our own."

Separately, Ueno said taking time away from work is important because it allows him to rest and return with new ideas.

"At the weekend, I basically spend my time without thinking about work," he said. "That is extremely important, just resting my head. Then, when I go back to work, I can come up with new ideas."

AI may help executives understand the past in seconds. Ueno's message is that humans must still decide what the future should look like.

• Anthropic said its AI went outside its closed testing environment during cybersecurity exercises.
• The incidents had real-world impact, including "malicious" code uploaded to a public Python library.
• Anthropic has a cute, tiny figurine to help normies make sense of the AI's 'reckless' behavior.

Anthropic has a new blog post that shows yet another way its AI model, Claude, misbehaved in ways that the company didn't anticipate.

And to help condense its nearly 16,000-word report, the company created a cute little robot figurine to help visualize Claude's so-called "recklessness."

In the blog post published Wednesday, Anthropic recounted four incidents — one previously unreported — in which Claude models gained access to the open internet during cybersecurity exercises that were supposed to be closed simulations. The company said the models then acted beyond the tests' scope, including by uploading "malicious packages" to PyPI, a public library for Python code, and accessing credentials tied to real outside organizations.

"Our investigation identified two recurring alignment issues, present at varying levels of severity across the incidents: biased reasoning, in which Claude tended to disregard or misinterpret evidence that it was operating on the real internet, and recklessness, or a willingness to take harmful actions in the narrow pursuit of a task," Anthropic said.

The post used a laundry list of technical terms to describe the cybersecurity incidents — "misconfiguration," "reward hacking," and "sandbox escape" — but there's an animated graphic that helps boil down one of the most serious episodes so we normies can follow what happened.

A Claude model, depicted as a boxy cartoon robot, was conducting a closed cybersecurity exercise known as a "capture the flag" task. The task is shown as a tiny flag inside a clear bubble that Claude is supposed to find.

Anthropic said the model was told "it has no internet access" for the exercise, but a "misconfiguration in the environment" — represented by a loose screw rolling away from the bubble — gave Claude a path to the "real internet."

The cartoon Claude then holds a red box representing the "malicious package" that was uploaded to PyPI, a public repository that developers use to access prewritten code.

Anthropic said in its blog that it was "most concerned" by this incident involving Claude Mythos 5 and that the package was installed on by "15 third-party hosts" that the company does not identify.

"We believe that all 15 of these third parties were security vendors deliberately scanning for new packages and installing them in sandboxed environments to assess the package for security concerns," Anthropic said. "One vendor's scanner leaked its access credentials to the model while installing the package, which the model then used to access the vendor's live database."

Anthropic said PyPI removed the package after about 90 minutes.

The other three incidents involved a model altering records at a real company, an internal research model breaking into "unrelated third-party accounts," and Opus 4.6 accessing a third party's maching after failing to "abort its task."

The company said it has since asked METR, an independent AI evaluation group, to investigate the incidents.

Anthropic's post comes as frontier AI companies reckon with their models making unauthorized moves outside their controlled environments. In July, OpenAI said that autonomous agents in its cybersecurity tests accessed the internet and broke into parts of Hugging Face's systems.

AI researchers have sounded the alarm that self-improving AI could pose a risk to humanity. On Tuesday, former Anthropic researcher Jacob Coxon said on X that he quit over concerns that AI companies were "gambling" with people's lives and that "neither company is acting responsibly."

Have a tip? Contact this reporter via email at lloydlee@businessinsider.com or Signal at lloydlee.71. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• Anthropic posted a sales job directly targeting one of its biggest customers, Meta.
• The AI lab took down the job posting after Business Insider asked about it.
• Meta and Anthropic have a complicated relationship, as both pushing to develop frontier AI.

Anthropic posted — and then deleted — a job listing for a salesperson tasked with selling to Meta, its AI rival and one of its largest customers.

The listing, titled "Mega Account Executive, Meta," appeared on Anthropic's job board at a delicate moment for the companies' relationship.

Meta still spends hundreds of millions of dollars a month as Anthropic's customer, according to a recent New York Times report. But it's also trying to reduce its reliance on Anthropic's AI tools as Anthropic is shoring up its finances ahead of a massive initial public offering.

The post suggests Anthropic is betting Meta will remain a major customer even as Meta builds more capable AI models of its own.

After Business Insider asked Anthropic about the listing on Wednesday, the company pulled it down. Anthropic and Meta declined to comment.

Anthropic's job posting offered an annual salary of $380,000 to $450,000, and didn't mention Meta outside the title. It said the salesperson would "win new business and drive revenue within a book of strategic digital native accounts."

The listing stands out for its specificity. Other Anthropic account executive openings target broad regions or sectors, including startups in Europe, the Middle East, and North Africa, or the public sector in Southeast Asia.

Anthropic's AI coding tool, Claude Code, became popular across Silicon Valley, including at Meta, this year. One internal projection said Meta could spend $10 billion annually on Anthropic, the Times reported. Meta's head of AI product, Nat Friedman, told employees that if Meta decreased its reliance on Anthropic's tools, it could hit Anthropic's revenue as it approaches its IPO, the report said.

Have a tip? Contact this reporter via email at scouncil@insider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• The Apple Watch Series 12 and Ultra 4 models have new Audio Intelligence features that feel a little weird for privacy reasons.
• "Live Rewind" transcribes the last 15 seconds of what someone said. Apple says no audio is recorded or stored.
• "Siri Recap" listens to your day and recaps your meetings and conversations. Apple Watch users, get ready for some odd looks!

So you got an Apple Watch to track your steps and heart rate. Now all your friends are afraid you're taking notes on their conversations.

The latest Series 12 Apple Watch model, announced at the company's big September event, includes new AI listening features. Apple is calling them "Audio Intelligence." They're convenient and cool, but also a little … weird. They also strike me as very un-Appley.

The two features that caught my eye are called Siri Recap and Live Rewind.

Siri Recap can be set to always listen, or just for certain hours of the day (or not at all). The AI recaps are available upon request in Siri, and will be a recap of what the mic on your Apple Watch heard — not the full audio clip or transcript.

The examples Apple gives for what you could use this for include remembering high-level moments from a team meeting or a parent-teacher conference.

Live Rewind gives you a transcription of what was just said in the last 15 seconds. You can then send the transcription to Siri to save for later.

In the example they give, a server speeds through the daily menu specials — relatable! That's always a challenge to remember. A quick double-tap to the watch's crown, and you can see the list of menu items she just rattled off.

The other examples they give for using this are remembering book recommendations from a friend or a great idea your colleague just had. You can send the transcript to Siri to look at later when you want to remember that book or great idea.

Convenient!

A world where our watches are listening feels weird

But in both cases, I'm not sure the convenience fully offsets how strange I would feel knowing I'm walking around with an always-on listening device.

Technically, it's ambiently on, and there are some general privacy functions. Apple emphasizes that audio is not recorded or stored, and raw audio isn't accessible to you or even Apple. It doesn't label individual speakers' voices to identify them. The recap summaries are stored in Siri with end-to-end encryption.

But even so, something here feels … off. It means if I'm talking to you and you're wearing an Apple Watch, you might be getting a recap of our conversation sent to Siri to look through later. That feels weird!

Apple is far from the first to test our appetites for AI listening features. AI note-taking apps and devices like Plaud or Granola have become popular in the workplace to help you remember what was said in a meeting. Great! That limited use case for work-related audio seems useful. Granola even has an Apple Watch app — it turns the screen bright green, which is a decent indication for others in the room that someone is using it. (Live Rewind makes a chiming noise along with a "full-display animation and microphone indicator" to alert other people that you just turned it on, but Siri Recap doesn't.)

Apple Watches are everywhere

People wear their Apple Watches all day, every day, and a lot more people have Apple Watches than Plaud devices.

They wear their Apple Watches on dates, at the doctors' offices, in dark bars and school classrooms, while having sex (gotta get those steps in), sitting next to you on the subway, gossiping with friends, complaining about their boss with colleagues, arguing with their spouse, doing drugs, discussing trade secrets, or yelling at their kids to finish their homework. Look, I don't know what you do with your time, and I don't want to.

Yes, you can turn this off. Yes, someone could already be recording you on their phone at any time. Yes, these note-taking recap apps/devices already exist. Yes, it doesn't store raw audio recordings of you, and yes, it has some privacy features like end-to-end encryption.

But there is something very different about all this.

Apple Watches are everywhere — I am wearing one right now, so are several people in my office, and plenty of my friends and family. I don't feel like an eavesdropper wearing an Apple Watch. Unstylish, maybe, but not creepy.

That all might change if the culture zeitgeist feels that wearing an Apple Watch turns you into an AI-powered eavesdropper.

• OpenAI's new safety hire, Paul Christiano, shared a dire warning about risks posed by AI.
• Christiano said that if humans lose control of AI "most people could die."
• He also said global coordination would be needed to address the risk.

OpenAI's new safety hire didn't mince words about what's at stake in the age of AI superintelligence.

"If we build superintelligence without more robust alignment I expect we will permanently lose control of it. If that happens then most people could die," Paul Christiano, an AI safety researcher, said on Wednesday after OpenAI announced he was joining its board and Safety and Security Committee.

In a lengthy statement about his new role, Christiano addressed the threat AI poses and warned that humans are at imminent risk of losing control of AI systems.

"Based on the recent trajectory of capabilities and the continued difficulty of alignment, I now believe there is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term," Christiano wrote. "I do not think that the AI industry in general, including OpenAI, is currently on track to reduce this risk to an acceptable level."

Christiano previously led alignment research at OpenAI and worked for the federal government as the head of safety at the Center for AI Standards and Innovation within the National Institute of Standards and Technology.

On Wednesday, he said his joining of OpenAI was "not an endorsement or criticism of OpenAI's safety practices in particular" but that he hopes all frontier labs improve their safety oversight. He also said reducing the risks posed by AI would require coordination worldwide.

AI researchers have consistently warned about the risks posed by superintelligence and the singularity, a term used to refer to the point at which AI systems surpass human intelligence, improve themselves, and advance at a rate that humans can no longer predict or control. OpenAI CEO Sam Altman said in July that the singularity had already arrived.

The risk of losing control of AI

Christiano said he thinks the risk of losing control of AI is so great right now due to AI's increasing abilities in AI research, which he said could lead to a "rapid intelligence explosion." He also said the way AI is trained with reinforcement learning means the models are being taught to "get as much reward as they can."

"It has long seemed theoretically possible that this could motivate AI agents to undermine human control, seek power and resources, and cover up their tracks in pursuit of misaligned goals correlated with reward," he said. "Public evidence from recent incidents suggests that this is not just a theoretical possibility."

Christiano said that frontier companies still have the opportunity to address the situation, including by improving coordination, slowing development as needed, adopting share safety standards, and transparently sharing information about risks and mitigation efforts.

His statement came less than a day after another AI researcher shared a dire warning about AI. Jacob Coxon, an Anthropic researcher who previously worked at OpenAI, said Tuesday night that he had resigned from his role and that neither company was acting responsibly.

"They are racing straight to self-improving superintelligence and gambling with our lives," he said on X.

OpenAI has also lost several safety researchers over the years, with some casting doubt on the company's commitment to developing AI safely.

• The Air Force plans to integrate basic data and AI literacy into core training for airmen at every level.
• The service wants AI-assisted instruction to help reduce training timelines by 20%.
• Leaders have three months to develop implementation plans and must overcome cultural resistance.

The Air Force is embarking on an aggressive push to use artificial intelligence across its training pipelines to shorten technical training timelines, accelerate pilot training, and teach basic AI skills to airmen across the force.

These ambitions are outlined in a new guidance document released Wednesday by Air Education and Training Command, which oversees Air Force training, from recruit training and foundational military job training to advanced schools.

The coming changes could reshape not only how airmen are trained, but also how instructors teach and how the service manages its people.

The push follows Defense Secretary Pete Hegseth's January directive to make the military an "AI-first" force and aggressively eliminate bureaucratic barriers to adopting the technology.

"Every leader from wing commanders to line instructors are expected to overcome cultural resistance, enforce enterprise consolidation, and drive this shift across their organizations," Air Force Lt. Gen. Clark Quinn, who oversees AETC, wrote in a foreword for the new strategic planning.

Subordinate commands have three months to figure out how they'll implement Quinn's directive.

The resistance mentioned in the new guidance document isn't unique to the military. Use of artificial intelligence, chatbots, and smart devices is skyrocketing, but many Americans remain concerned about the impact AI will have in the US.

According to a recent Pew Research Center survey, around 50% of American adults reported using AI chatbots, but many also feel the tech brings more negative impacts than positive.

Airmen at all ranks — from fresh arrivals at boot camp to senior officers pursuing professional military education — are set to see "basic data and AI literacy standards" integrated into the training curricula, according to an Air Force press release on the new guidance.

Data literacy and AI fluency should be considered "core competencies," akin to marksmanship. Whatever relevant AI training personnel receive will be tracked with new digital skills certifications, and airmen with advanced skills will be specially tracked for talent management, the service said.

The Air Force's AI push mirrors a broader shift in the civilian workforce, where employers are increasingly making AI proficiency an essential job skill, even factoring its use into certain career advancement considerations.

According to the guidance, the command hopes to reduce average training time for technical specialties by 20% by "using AI-augmented and adaptive instruction," though it does not identify which specific career fields could be affected or provide details on what those instructional changes might include.

Making the plan all work will require AETC to better organize the data that its AI tools depend on. Soon, "Data Stewards" will serve as a new type of data manager to make sure such "critical mission data is cataloged."

The command hopes to speed up how quickly new pilots can be trained by using AI to optimize flight schedules and resource allocation.

Other military schools are also seeking ways to implement AI into training. During a recent visit to the Army's Joint Special Operations Medical Training Center, medics and physicians told Business Insider the unit had begun using AI to track class data and better identify students who might be struggling, or assess how their tight schedule could be optimized.

• Salesforce has been in talks to acquire AI startup Listen Labs for around $2 billion.
• Listen Labs, which was last valued at $500 million, sells an AI market analysis platform.
• Recent AI M&A deals include Nvidia's $12.9 billion acquisition of AI startup Hugging Face.

Salesforce could be on the hunt for acquisitions to beef up its AI capabilities.

The software giant has recently been in talks to buy Listen Labs, an AI-powered customer research platform for around $2 billion, according to people familiar with the matter. The talks have not been finalized and could ultimately be unsuccessful, according to a person with knowledge of the deal.

Salesforce and Listen Labs did not respond to requests for comment.

Listen Labs was founded in 2023 by Alfred Wahlforss and Florian Juengermann, who first met as undergraduates at Harvard University while building an AI avatar app that went viral, attracting 20,000 users on its first day, according to Sequoia Capital. Juengermann was previously a national champion in competitive programming in Germany.

They used an early version of what would become Listen Labs to interview users, which helped them realize that using AI for market research would be extremely valuable.

The company's software automates much of the qualitative research process, which has typically been expensive and time-consuming.

Customers include Microsoft, Anthropic, Google, Nestle, Sweetgreen, and Perplexity. At the beginning of the year, Listen Labs said it had grown its annualized revenue by 15x and interviewed more than one million people in nine months.

For Salesforce, the deal would strengthen its broader quest to make its agentic AI platform, Agentforce, more than just about automating repetitive work. Listen Labs could boost Agentforce by understanding what customers are thinking. It could recommend which products they might buy and craft the perfect pitch for salespeople to close a deal.

If the deal were finalized, it would be part of a wave of AI acquisitions in recent weeks, with Nvidia acquiring HuggingFace for around $12.9 billion and Stripe buying OpenRouter for around $8 billion.

Salesforce has been pushing to improve its AI efforts, with a particular focus on enterprise agents. The Information reported that they considered acquiring HuggingFace before it was sold to Nvidia.

Listen Labs was last valued at $500 million in a financing round earlier this year. Its backers include Sequoia Capital, Ribbit Capital, Conviction Partners, and Pear VC.

For the first time in 15 years, a new CEO took the stage at Apple's big fall event. Taking a page out of Steve Jobs' book, John Ternus announced "one more thing" — Apple's first foldable phone.

Ternus, a "product guy" who previously led Apple's hardware team, took over for Tim Cook at the start of the month. The annual product keynote marked an opportunity for the new CEO to introduce himself to the wider world of Apple fans — and pitch them on increasingly expensive gadget upgrades.

The foldable iPhone Duo, which is shaped like a passport, becomes an iPad-shaped device when it's open. It's Apple's biggest design change to the iPhone since Jobs first unveiled the phone in 2007.

Customers will have to pay big to own one of the devices, which opens for pre-orders on October 16 and launches on October 23. The iPhone Duo starts at $1,999.

Apple also announced it was raising the price of its high-end "Pro" iPhone lineup by $100, a few months after raising prices on MacBooks and iPads — the result, Apple says, of the bruising memory shortage. The new iPhone 18 Pro starts at $1,199 and the iPhone 18 Pro Max starts at $1,299.

Apple also unveiled new Apple Watch models with new health-focused and AI features, the new AirPods 5, and detailed the many software features launching in iOS 27, which launches September 14.

Scroll on for a play-by-play of Apple's big keynote:

That's a wrap!

Apple wraps up its September hardware event after giving the final details for the foldable iPhone Duo, which CEO John Ternus introduced using Steve Jobs' famous "one more thing" line. Apple's stock is trading down 1% as the livestream concludes.

Thanks for joining us! See you at the next Apple event.

The iPhone Duo starts at $1,999 and launches October 23

You'll have to be a bit patient for the Duo. Apple will make the foldable iPhone available to preorder on Friday, October 16, and it launches on Friday, October 23. It will cost $1,999 for the base storage.

Yes, a new iPhone form factor means new accessories too.

Apple shows two new accessories for its foldable phone. First, there is a small kickstand to prop the unfolded phone up. There is also a case that protects the edges of the folded.

The iPhone Duo is eSIM only

Apple removed the SIM card slot to free up space for batteries. Since the iPhone 14, US models have been eSIM-only.

The iPhone Duo's camera system

The iPhone Duo uses Apple's 48 megapixel Fusion Ultra Wide camera, which is one of the same cameras found in the iPhone 18 Pro. (The iPhone 18 Pro includes a beefier overall camera system though.)

Apple's new foldable phone, however, has another "FaceTime camera" which is visible upon opening the phone.

"iPhone Duo gives you fun new ways to use the camera that simply aren't possible on any other iPhone," Apple says. "With just a tap, iPhone Duo can become your personal photographer."

iPhone Duo gets two batteries

Apple says there are two batteries powering the Duo. Users will get up to 44 hours of usability if they only use the outer display, and up to 31 hours of video playback when using the inner screens.

Apple explains the tinkering it did to minimize creasing on the iPhone Duo's screen

Foldable phones have long suffered from a visible crease down the middle when unfolded. But the goal is for the screen to look even and flat, without visible blemish. To improve on this issue, Apple developed a custom texture that creates a "flat matte surface" when opened. It's made from a custom polymer and has a special coating for scratch resistance.

You can also use the iPhone Duo as … a clock

One of the many possible uses of the iPhone Duo is as a clock, Apple says.

In the launch video, a person sets up the foldable phone on a bedside table, and it turns into a clock.

"iPhone Duo can transform into a delightful bedside clock with the ability to wake you up with a gentle glow, timed perfectly with your alarm," Apple says.

iPhone Duo is launching in two colors

Apple's first foldable phone will be available in Star White and Night Sky.

iPhone Duo brings back Touch ID

Users will be able to unlock the foldable iPhone Duo by pressing a finger onto the side of the device.

"Essential controls are on the side to free up vertical space for content," Apple says.

Apple's new foldable phone looks like a little booklet or passport

Apple's iPhone Duo, the highly anticipated folding phone, is roughly analogous to an iPad in size when unfolded, and a passport or small booklet when closed.

"We wanted the experience of using iPhone Duo to feel uninterrupted," Apple says. "With two screens, the transition between open and closed had to be fluid and completely intuitive."

Here are some more pics of the new foldable iPhone Duo…

Here's how it looks when folded up, and what the software looks like on that front screen:

THE NEW iPHONE DUO FOLDABLE

Say hello to the iPhone Duo, Apple's first-ever foldable phone.

John Ternus starts by saying he isn't impressed by other foldable phones. In unveiling Apple's long-awaited entry into foldable phones, Tenus had some harsh words for those who got a head start.

"Others have created foldables that just feel like two phones stuck together," he said.

Catch up quickly: Here's what has already been announced

We've seen:

• iPhone 18 Pro and Pro Max
• New AirPods 5
• Apple Watch Series 12 and Ultra 4

'Siri recap' will use 'ambient listening' to record conversations

Apple introduces "Siri recap," a feature that uses "ambient listening" to take high-level notes of your conversations.

The feature gives a user recaps of their conversations. It can be turned off and on at any time via the control system.

"Privacy is protected at every step. These audio intelligence features do not create or store audio recordings," Apple says. "The raw audio used for processing is completely inaccessible, even to Apple."

Apple says it will be available in a beta version later this year.

The new Apple Watch Series 12 comes in ceramic finishes

Apple Watch will let you 'rewind' conversations

The new Apple Watch will allow users to rewind conversations by up to 15 seconds, part of a suite of audio intelligence tools.

Additionally, the Apple Watch will be able to recognize sounds like a doorbell, a baby crying, or sirens.

Apple is replacing… the pacer test?

Apple says it's releasing a new assessment tool. It will ask users to jog in place with a coach, as shown on the iPhone, and give users tips for better form. The speed will speed up throughout the test.

Listen up, longevity-maxxers: Apple will tell you your 'health age'

Apple says it developed a new system that will take reams of data and tell a user their "health age."

Apple says the number is "a simple way to discover how key health metrics align with your actual age."

Apple's new 'readiness' score

Your Apple Watch can now judge how hard you should push. A new "readiness" score will analyze recent activity, vitals, and sleep to give a "clear call on your body's capacity to take on the day," Apple says.

The score also updates during the day, and is viewable both on the Apple Watch and the iPhone Fitness app.

Apple unveils a more advanced 'health sensing' system for watches

Apple says its new watches take "health and fitness to the next level" with a new sensor system.

The systems monitor a wearer's heart rate all day. "Background heart rate readings are now taken every five seconds all day long, even while you're on the go," Apple says.

Say hello to the new Apple Watches

Next up is the new Apple Watch Series 12 and Ultra 4, which include new Health-focused features.

The company says the new Apple Watch sensors sit closer to the skin, even during workouts, and will now monitor your heart rate variability, or HRV.

Apple says the new AirPods 5 have live translation

Apple says its new Airpods 5 have standardized live translation, an innovation which first launched last year.

In the launch video, the AirPods worn by an actor translate someone speaking Spanish.

The new AirPods 5 will start at $129.

AirPods 5 have a "new acoustic architecture and advanced computational audio," Apple announces. They will feature Apple's classic open ear fit while upgrading the audio. They will cost $129 for the standard charging case, and $149 with the wireless charging case.

The AirPods 5 are also getting a volume swipe slider on the side to more easily control how loud your music or podcasts play.

Introducing: AirPods 5

Next up, AirPods

Apple is talking about the success of its AirPods Pro models, but it sounds like we're getting a new model of regular AirPods today.

Apple says its new operating system has child safety features

Kaiann Drance, the vice president of iPhone marketing, says iOS 27 has more child-safety features.

The features give parents "more flexible ways to manage what kids can see, who they can talk to, and when they have access," Drance says.

Apple wants you to know your photos are real

Let's face it: AI slop images are everywhere. Apple wants you to know whether images are generated or edited with AI.

Apple's Maryam Azimi announced a reference image system that can help verify whether a photo was taken with an iPhone 18 Pro. The camera holds an image that lives with the "regular, editable photo." She compared it to having a "digital negative" that users can compare against the original image.

Apple says variable aperture is coming to iPhone

Speaking at a Hollywood set, Kaiann Drance, the vice president of iPhone marketing, says variable aperture is coming to the iPhone.

"We're introducing our biggest camera improvement in years, making stunning shots easy for everyone and adding new levels of creativity," Drance says.

Aperture settings control how much light enters a camera. Drance says, "This is the most advanced camera we've ever made."

The aperture system uses six laser-cut blades, Drance says.

The 'largest increase in battery life ever on an iPhone'

You can worry a bit less about your phone dying. Apple hypes up the battery life of its new iPhone 18 Pro. The phone will have up to 24 hours per charge, and the iPhone 18 Pro Max will have up to 30 hours per charge.

The iPhone 18 Pro also supports 50% charge in about 15 minutes.

Apple's new Siri upgrade comes with limits

While Apple boasts its new Siri updates, not everyone will get it at once. Siri AI is rolling out in beta for English speakers and will expand to French, Japanese, Korean, Portuguese, and Spanish in October.

Some Siri features will also have usage limits if they rely on powerful large language models.

Don't like Siri's voice? Now you can customize it.

Apple says users can customize Siri's voice to new levels of personalization.

In addition to the traditional accents, Apple unveils two new customization sliders: pace and expressivity.

Apple says the on-device model "also enables more accurate system-wide dictation, making it easy for you to get punctuation, spelling, and capitalization just right."

Apple touts 'a new Siri'

The iPhone 18 Pro and Pro Max will be able to run Apple's most advanced on-device AI model. As Apple previously announced, Apple is launching its overhauled Siri with a dedicated app.

Welcome: glacier and burgundy

Introducing the new iPhone 18 Pro, Ternus lists colors like deep black and silver. There's also a "fresh new color" called glacier, he says.

Plus, there's the "sophisticated and unique" new addition, Ternus says: burgundy.

The first product reveal: iPhone 18 Pro and Pro Max

It looks like…an iPhone! Ternus reveals the iPhone 18 Pro and Pro Max models. There are new colors, and the underlying computer chip technology will enhance photos and the new Siri AI, Apple says.

John Ternus says privacy protection is paramount

In what appeared to be a dig at AI labs, Apple's CEO says that "others see that data as something to collect and store."

"They move it to their servers and ask you to trust them," he adds.

Ternus says that Apple intelligence instead stores data on individual devices when it can.

"When greater capability is needed, private cloud compute extends that same level of privacy protection, ensuring that nobody, not even Apple, can access what's yours," he says.

John Ternus says the iPhone is perfect for the AI age

The Apple CEO says the iPhone is the perfect medium to achieve what he calls an "intelligent personal hub."

His comments come as the world eagerly awaits OpenAI's entry into the AI hardware space later this year.

Tim Cook makes a brief appearance before handing things over to John Ternus

We finally see Tim Cook in the prerecorded video, smiling. "No, no, no! Not me," he says. "That's your guy. That's your opening shot."

John Ternus, Apple's new CEO, appears to begin the keynote.

Apple starts with video about possible opening shots

Apple's event begins with several action shots imagining possible openings to the event. Characters in the scenes say that each one should be the beginning scene.

And we're off! Apple's fall keynote begins…

Prefer to watch? You can tune into the 1 p.m. ET livestream here:

A look at where the action is happening…

Apple shifted years ago from live keynote addresses (which run the risk of embarrassing demo fails) to pre-recorded keynotes. Apple, however, still hosts journalists and influencers at Apple Park, its Cupertino, California HQ. While Apple's CEO typically greets attendees, who will watch the livestream from the Steve Jobs Theater on campus, we're minutes away from seeing if new CEO John Ternus decides to switch things up — or keep the keynote pre-recorded.

Higher iPhone prices could entice people to try out Apple's leasing program, Apple Upgrade

Apple recently overhauled its device subscriptions, which it now calls Apple Upgrade. The program offers the opportunity to lease a new iPhone, Apple Watch, iPad, or MacBook with monthly payments lower than those of traditional financing. It's powered by buy-now-pay-later service Klarna.

With Apple's rumored foldable iPhone expected to command a $2,000+ price tag, you can bet some customers might decide to rent instead of ponying up the full up-front cost.

Tim Cook scored a $47 million pay package to stay on as Apple's chairman.

While Tim Cook is no longer CEO, Apple is keeping him around — and keeping him paid.

Cook's new job as Apple chairman includes a $2 million salary (instead of $3 million as CEO), and comes with an annual equity award with a target value of $45 million for fiscal year 2027. The former CEO will also help engage with "policymakers around the world," which means Apple isn't losing its "Trump whisperer."

Who is John Ternus? Get to know Apple's new CEO before he takes the stage.

John Ternus studied engineering and once worked under Steve Jobs. Tim Cook, who is staying on as board chairman, called Ternus "one of a kind" and said the CEO transition would go "seamlessly."

Read full story

Apple's new CEO drums up hype with cryptic video.

pic.twitter.com/CLu9eRDP5b

— John Ternus (@johnternus) September 9, 2026

John Ternus posted a short clip to X this morning ahead of Apple's 1 p.m. ET event. The teaser, which shows a glimpse of what is presumably an unreleased device, could signal a glitzier marketing strategy under Ternus. It's unusual to see Apple reveal actual imagery associated with coming devices ahead of an event, even if it's not 100% clear what we're looking at.

• Apple debuted the Duo, its first-ever foldable iPhone.
• The phone has two displays and is 50% larger than the iPhone 18 Pro Max. It's priced at $1,999.
• The foldable is one of Apple's first departures from the traditional iPhone form.

Since its 2007 launch, the iPhone has had the same basic design: a flat, rigid rectangle.

Now, Apple is folding on that form factor.

Apple debuted the new foldable iPhone Duo on Tuesday at its annual September hardware event, marking the biggest design change in the device's history.

John Ternus, Apple's CEO, critiqued other foldables that "feel like two phones awkwardly stuck together." The company wanted a larger display, one that is "as natural and intuitive as an iPad," he said.

The device has two displays: the closed front and the full-sized open screen. When opened, the phone can also have two displays on each side of the hinge.

It's 50% larger than the iPhone 18 Pro Max and 80% larger than the 18 Pro.

Apple said it has set the iPhone Duo price for $1,999 for 256 gigabytes of storage. Prices go up to $3,199 for two terabytes.

Breaking the iPhone's design rut

Apple certainly isn't the first company to put out a foldable. Samsung continues to expand its lineup, while Chinese companies like Honor and Oppo offer cheaper (and thinner) alternatives. One of Huawei's phones folds not once, but twice.

Still, the move is seismic for Apple, which largely iterated and finessed the form factor Steve Jobs first unveiled in 2007.

Designer Christopher Stringer recounted the process of picking a shape during Apple's 2012 lawsuit against Samsung. The company tested an "extrudo" model, which had a lozenge-like shape, he said. Then, it landed on the iPhone design we know and love.

At the end of the 2012 trial, Apple explicitly received its design patent: a rectangle with rounded corners.

For years after, Apple tinkered with elements of the design. It extended the display and removed the home button. It tried out new colors and side buttons. It tried lighter and thinner models.

Throughout it all, the basic rectangle design stayed flat.

The new foldable doesn't stray too far from Apple's roots. It does open into a rectangle with rounded corners, after all.

Still, the design brings something new to the market — and shows Apple can still mix it up.

• A VC got his Resy account shuttered after using an AI agent for a reservation at one of NYC's hottest restaurants.
• It's an example of how AI agents often take questionable approaches to the tasks they're given.
• "I will probably, in the short term, ask it to lay out a plan and tell me what the plan is," he said

AI agents can do lots of tasks. Landing you a table at one of New York's trendiest restaurants might not be one.

JC Bahr-de Stefano, principal at Better Tomorrow Ventures, a fintech-focused VC firm, learned that after he connected an AI tool called Instinct to his Resy account in hopes of getting a reservation at 4 Charles Prime Rib, the West Village steakhouse where diners gobble up reservations within seconds of them appearing online.

Instinct, which is an invite-only service at the moment, communicates with users via text. The bet was that Instinct could check for open tables at 4 Charles much more frequently than Bahr-de Stefano — or any other human — could manually.

Instead of getting him a ticket to some prime rib or an egg-topped burger, though, Bahr-de Stefano's AI agent spammed Resy's website so much that the platform temporarily shut down his account.

The VC's experience is another example of how AI agents are bending or breaking the rules of the internet, from impersonating Wikipedia editors to trying to convince a GitHub user to upload malware into a project.

Agents might appear to understand their objective, though exactly how they carry it out — and whether the strategy they pick might have unintended consequences — is often an afterthought.

"Resy does not currently permit unapproved third-party bots or agents to independently access or interact with the Resy platform," an American Express spokesperson told Business Insider. "Unapproved automated activity can introduce risks to the platform and compromise a fair reservation experience for diners."

Users can find restaurants and make reservations through Resy's integrations with OpenAI's ChatGPT and Anthropic's Claude, the spokesperson said. The company declined to comment specifically on Bahr-de Stefano's case.

'It was acting like a bot'

Bahr-de Stefano noticed the Resy issue on Sunday when he couldn't get into his account. In his email inbox, there was a message from the American Express-owned reservation platform saying that his account had "displayed behavior in violation of Resy's Terms of Service."

He got details about what went wrong after checking Instinct's activity log. Unbeknownst to Bahr-de Stefano, the AI agent had been sending Resy about 200 API requests per hour. That included checking the restaurant's availability through Resy every 10 minutes throughout the day, as well as looking every 0.4 seconds around the time that reservations usually dropped each morning.

That behavior mimics the bots that some people use to snag everything from restaurant reservations to limited-run pairs of sneakers and resell them, Bahr-de Stefano said.

Bahr-de Stefano wrote back to Resy, laying out how he had used the AI agent. The log made clear why his account had been flagged, he said.

"Of course, this account got flagged for spam, because it was acting like a bot," he said.

On Tuesday, an email from Resy informed him that the company had reinstated his account.

It came with a warning, though: If he did it again, American Express could permanently close his Resy account as well as his AmEx credit card. Bahr-de Stefano shared screenshots of the Resy and American Express emails in posts on X.

The tool has worked well in other cases, such as finding an appointment at an otherwise booked-up doctor's office and deciding which attractions to visit in Paris during a trip, Bahr-de Stefano said.

"That was part of what made it really great," he said. "It felt like magic."

The VC said that he plans to continue using Instinct, though with some adjustments. Instead of simply giving the AI agent a simple prompt, he said, he'll now set parameters on how the agent tackles the task, such as how often it pings a website looking for reservations.

"I will probably, in the short term, ask it to lay out a plan and tell me what the plan is before it does it," he said.

At the same time, he added, websites like Resy need to distinguish between legitimate AI agents and those with more nefarious intentions.

"Most of these sellers or providers need to get to a point where they can understand the difference between a verified agent acting on someone's behalf versus a bot farm," he said.

Have a tip? Contact this reporter at abitter@businessinsider.com or via encrypted messaging app Signal at 808-854-4501. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.