Gemini Spark is Google’s take on a steroided-out assistant agent that knows everything about you, announced as part of the company’s updates to its Gemini chatbot app at this year’s I/O developer conference.

Software companies have been talking up AI agents for some time now, but I wasn’t impressed until I tried Anthropic’s Claude Cowork in January. I sat back as the bot organized the scattered screenshots littering my desktop into labeled folders without a single click, and felt convinced that this might be a turning point for how people interact with their computers.

Many other early adopters in San Francisco experienced similar moments when they set up the mega-viral OpenClaw bot earlier this year, not just to help complete a few tasks but to run their whole online lives. Power users attempted to fully automate their inboxes, calendars, and text messages, and even run a vending machine to varying levels of success via OpenClaw. It’s not without risks—you have to give these agents control of your data and computer, and OpenClaw almost deleted an entire trove of emails for one Meta employee who was experimenting with it

Whether it’s my daily schedule via Google Calendar or my date-night dinner spots through Gmail confirmations, Gemini Spark can dive deep into the well of my personal info before I even connect to a third-party integration. While the standard Gemini app can complete many of the same tasks, Sparks’ differentiator is that it proactively gathers details and takes action while you’re away, rather than waiting for you to prompt it.

Google pitches Gemini Spark as a one-stop shop for completing tasks people previously handled manually or did in other apps. The agent can look through your credit card bill regularly to flag surprise fees—sorry, RocketMoney app, won’t be needing you anymore. Spark can be calibrated to automatically skim every email about your preschooler and highlight key dates for a morning digest report. You can even throw all your meeting notes at Spark and ask it to draft a Google Doc and generate follow-up emails to the right people.

This agent is getting a slow rollout, arriving for a small group of early testers this week and launching next week in beta for subscribers to Google’s $100+ per month AI plan. It’s pricey to be one of the first people to experiment with Spark! The company plans to allow Spark to connect through Gemini to third-party apps, like OpenTable and Instacart, for additional automation opportunities in the coming weeks. Other features imminent on the Spark road map include allowing the agent to manipulate your local browser and the ability to text or email commands to the agent.

Being able to text commands to your agent sounds like a key factor in actually making the Spark experience feel seamless. Rather than opening the Gemini app and getting distracted, I’ll spend all day texting Spark my increasingly niche requests, as if it were assistant Andrea from The Devil Wears Prada.

One of the main measures of success when trying this agent will be how often it goes off the rails. “Spark operates under your direction,” reads Google’s announcement blog about the agent. “You choose whether to turn it on and what apps it connects to, and it’s designed to ask you first before performing high-stakes actions like spending money or sending emails.” Anyone who tries the tool is taking a risk by using experimental software that’s powered by personal data.

Google plans to expand the agentic shopping feature to allow users to set spending limits and preferred merchants that Spark will adhere to, though exercising caution is critical. “We think of it as if you’re giving a teenager their first debit card,” says Josh Woodward, vice president of Google Labs and the head of the Gemini app.

Much like the changes Google is implementing in Search, which brings agentic task automation without needing to leave the search experience, Spark is Google’s chance to push AI agents further into the public zeitgeist. Let’s see if it has the necessary spark to pull it off.

Two former OpenAI employees and a group of AI safety nonprofits are warning that Elon Musk’s AI lab, xAI, could become a liability for prospective investors in SpaceX, which is preparing to file what’s expected to be the largest initial public offering in Wall Street history.

In a letter directed to investors published on Tuesday, the ex-staffers highlighted what they describe as “unpriced risks” related to xAI that could complicate SpaceX’s reported plans to raise up to $75 billion as part of its IPO. The rocket company’s private valuation shot up to over $1 trillion after it acquired xAI last year. Musk claimed his rocket company could launch data centers into space for his AI lab, but the letter’s authors argue that xAI’s poor record on safety issues could complicate how investors view the combined company as it gets ready to submit its IPO prospectus filing.

One of the letter’s signatories and coauthors is a new nonprofit called Guidelight AI Standards, which was cofounded by former OpenAI safety researcher Steven Adler and former OpenAI policy adviser Page Hedley. The group, which is backed by private donors, aims to improve the safety practices of frontier AI companies. Other AI safety nonprofits also signed on, including Legal Advocates for Safe Science and Technology, Encode AI, and The Midas Project.

Hedley tells WIRED in an interview that he believes xAI has the worst safety practices “nearly across the board” compared to other frontier AI developers, including OpenAI, Google DeepMind, and Anthropic. As a result, he argues, SpaceX may face a greater risk of regulation and litigation than other AI labs.

The letter’s authors argue that SpaceX should make several disclosures to investors, including whether xAI intends to continue developing frontier AI models. SpaceX recently struck a deal to sell a significant portion of its GPU capacity to Anthropic, and the letter claims the agreement “leaves it unclear whether xAI is still a frontier-AI competitor inside a larger holding company.” If xAI continues to develop frontier AI models, the authors say, it should be required to publish a public safety and governance plan.

SpaceX and xAI did not immediately respond to WIRED’s request for comment.

The letter also outlines examples of how xAI has not kept up with industry-standard safety practices such as publishing detailed frameworks for mitigating risks around its AI models being used in cyber attacks. The authors also outline specific safety incidents at xAI that they say warrant additional scrutiny. Among the most notable include when xAI’s flagship AI chatbot, Grok, spontaneously brought up white genocide in its responses. In another case, xAI allowed Grok to generate thousands of sexualized images of women and children, which spread widely across Musk’s social media platform X. The latter case prompted at least 37 US attorneys general to send a letter demanding that Musk’s AI lab take steps to protect women and children on its platform.

Hedley says the number of safety incidents xAI has experienced and the regulatory attention they received is “far out of proportion to its market share.” As lawmakers grow increasingly alarmed by the cyber capabilities of advanced AI models like Anthropic’s Claude Mythos, new security regulations may be on the horizon. The Trump administration is reportedly already weighing an executive order that would give US intelligence agencies more oversight over AI models.

“It takes serious investment to rein in [AI safety] risks, and it seems that xAI has historically under-invested here,” says Adler. The letter cites reporting from The Washington Post that said xAI had just “two or three” people working on safety as of January. “A question investors should be wondering is if xAI stays at the frontier, how costly might it be to, in fact, manage these [risks] responsibly? If they don’t, what might be the consequences?”

Meta CEO Mark Zuckerberg and his wife Priscilla Chan are hiring a seasonal, on-call “Beach Water Person” based in Kauai, Hawaii, where the family owns a sprawling compound, according to a new job listing on Greenhouse associated with West 10, the Zuckerberg family office.

This is an interesting choice for a job title, because according to the job description, the primary duties of this “Beach Water Person” include serving as a “Beach Lifeguard,” and “Pool Lifeguard.” In other words, being a lifeguard.

The job listing names a few additional duties related to water activities, such as instructing “stand-up paddleboarding (SUP), canoe paddling, snorkeling, and other ocean-based activities.” These, however, come after the water safety duties in the job description.

This position easily could have been called “Pool/Beach Lifeguard,” or simply “Lifeguard.” For the sake of comprehensiveness, “Pool/Beach Lifeguard and Boat Deckhand” would have also worked. Alternatively, the Zuckerbergs could have chosen “Beach/Pool Attendant,” a job title roughly synonymous with lifeguard that could reasonably be interpreted as encompassing extra duties associated with leisure, such as tending to a boat or teaching people how to stand-up paddleboard.

Arguably, any of these options would have provided more clarity than “Beach Water Person,” which does not appear to correspond with a job title anywhere else in the English-speaking world.

WIRED did not immediately hear back from representatives of the Zuckerberg family. Lacking a human to speak with, we decided to ask Meta’s AI chatbot “what is a ‘beach water person’?”

“‘Beach water person’ would just mean someone who loves being in/near the ocean,” the chatbot said. “The word for that is thalassophile—’a person who loves the seas and oceans.’” Ok!