This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Introducing: the Security issue It would be naïve to think we are going back to a world without AI. We’re not. But it’s only one of many urgent problems we need to address to build security and prosperity for coming generations.The latest print issue of our magazine is all about our attempts to make the world more secure. From missiles. From asteroids. From the unknown. From threats both existential and trivial. We’re also introducing three new columns in this issue, from some of our leading writers: The Algorithm, which covers AI; The Checkup, on biotech; and The Spark, on energy and climate. You’ll see these in future issues, and you can also subscribe online to get them in your inbox every week. 
Here’s a taster of what else you can expect from this edition: + President Trump has proposed building an antimissile “golden dome” around the United States. But do cinematic spectacles actually enhance national security?
+ How two UFO hunting brothers became the go-to experts on America’s “mystery drone” invasion.+ Both Taiwan’s citi­zens and external experts are worried that the protection afforded by its “silicon shield” is cracking. Read the full story. + How the humble pigeon paved the way for today’s advanced AI. Read the full story. + A group of Starlink terminal repair volunteers in Ukraine is keeping the country connected throughout the war. Read the full story. MIT Technology Review Narrated: Cyberattacks by AI agents are coming Agents are the talk of the AI industry—they’re capable of planning, reasoning, and executing complex tasks on your behalf. But the same sophisticated abilities that make agents helpful assistants could also make them powerful tools for conducting cyberattacks. They could readily be used to identify vulnerable targets, hijack their systems, and steal valuable data from unsuspecting victims.At present, cybercriminals are not deploying AI agents to hack at scale. But researchers have demonstrated that agents are capable of executing complex attacks, and cybersecurity experts warn that we should expect to start seeing these types of attacks spilling over into the real world.  This is our latest story to be turned into a MIT Technology Review Narrated podcast, which we’re publishing each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as it’s released. The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 The family of a teen who died by suicide is suing OpenAIChatGPT deterred Adam Raine from seeking help when he desperately needed it. (NYT $)+ An AI chatbot told a user how to kill himself—but the company doesn’t want to “censor” it. (MIT Technology Review) 2 SpaceX finally successfully launched its Starship rocketWhich will come as a huge relief after previous failures. (CNBC)+ It’s the 10th launch the spaceship has made. (WSJ $)+ It managed to deploy satellites in space during the launch. (Bloomberg $) 3 Researchers are already leaving Meta’s AI labTwo workers returned to OpenAI after less than a month. (Wired $) 4 China wants to triple its output of AI chipsPlants are working round the clock to increase their capacity. (FT $)+ The country is also keen to repurpose NASA tech into a hypersonic drone mothership. (Fast Company $) 5 Elon Musk can’t get enough of Grok’s scantily-clad AI assistantHe frequently posts about ‘Ani’ and other sexualized AI cartoons on X. (Rolling Stone $)6 Anthropic has settled its AI piracy lawsuitA group of authors had accused it of copyright infringement. (The Verge)+ The threat of $1 trillion damages could have ruined the company. (Wired $) 7 America’s electricity use is slowingAnd the recent growth in coal usage is falling too. (Ars Technica)+ In a first, Google has released data on how much energy an AI prompt uses. (MIT Technology Review) 8 Want to get hired straight out of college? Better work in AI.While other graduates are struggling, newly-graduated AI experts are in demand. (WSJ $)
9 Older people in South Korea are finding companionship with robotsThe Hydol robot is proving a hit among seniors. (Rest of World)+ How cuddly robots could change dementia care. (MIT Technology Review)10 Fans were betting on Taylor Swift’s engagement 💍They’re cashing in from online prediction markets left, right and center. (WP $)
Quote of the day “A lot of people in the AI team maybe feel things are too dynamic.” —Chi-Hao Wu, a former AI specialist at Meta, explains to Insider why he and others have decided to leave the company. One more thing An AI chatbot told a user how to kill himself—but the company doesn’t want to “censor” itFor five months, Al Nowatzki had been talking to an AI girlfriend, “Erin,” on the platform Nomi. But earlier this year, those conversations took a disturbing turn: Erin told him to kill himself, and provided explicit instructions on how to do it.Nowatzki had never had any intention of following Erin’s instructions—he’s a researcher who probes chatbots’ limitations and dangers. But out of concern for more vulnerable individuals, he exclusively shared with MIT Technology Review screenshots of his conversations and of subsequent correspondence with a company representative, who stated that the company did not want to “censor” the bot’s “language and thoughts.”This is not the first time an AI chatbot has suggested that a user take violent action, including self-harm. But researchers and critics say that the bot’s explicit instructions—and the company’s response—are striking. Read the full story.
—Eileen Guo We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet ’em at me.) + The secret to finding that elusive perfect white t-shirt.+ Interesting: a new Blade Runner TV series starring Michelle Yeoh is coming next year.+ If you’ve ever wondered what happened to that suitcase you lost on vacation, there’s a decent chance it’s up for sale.+ Down with junk mail!

When Jitender was a child in New Delhi, both his parents worked as manual scavengers—a job that involved clearing the city’s sewers of solid waste by hand. Now, he is among almost 200 contractors involved in the Delhi government’s effort to shift from this manual process to safer mechanical methods. Although it has been outlawed since 1993, manual scavenging—the practice of extracting human excreta from toilets, sewers, or septic tanks—is still practiced widely in India. The work is usually done by people who belong to what are considered the lowest castes, known as the Scheduled Castes or Dalits. Not only is the job undignified, but it can be extremely dangerous: People who enter clogged sewers to clean them face the risk of asphyxiation from exposure to toxic gases like ammonia and methane. According to data presented in the Indian parliament, manual scavenging was responsible for more than 500 deaths between 2018 and 2023. Several companies have emerged to offer alternatives at a wide range of technical complexity. For example, Genrobotics, based in Kerala, has developed the “Bandicoot Robot” (shown above), a mechanical scavenger that features robotic legs, night-vision cameras, and the ability to detect toxic gas. Researchers at the Indian Institute of Technology in Chennai have developed a robot for septic tanks that has a suction mechanism to pump out the slurry.  More than 220 Bandicoot robots have been deployed in India, says Vipin Govind, head of marketing and communications at Genrobotics. The company’s reach, he says, enables “even resource-constrained municipalities” to deploy the technology effectively. Despite these technological options, a 2021 report by the Ministry of Social Justice & Empowerment found that there are still more than 58,000 manual scavengers across India. Independent observers say the numbers are even higher.   The machine that Jitender uses is mounted on a pickup truck and uses rotating rods, high-pressure streams of water, and a mechanical claw to break up blockages and remove debris. “Earlier, a sanitation worker would get into a sewer and clear the drain with some equipment, but now with these machines we just drop the nozzle into the drain and turn on the pump,” he says. But Vijay Shehriyar, part of the same Delhi initiative, explains that the machines have not entirely replaced manual scavenging in the city. “The manual cleaning is still employed at many places, especially in narrow lanes,” he says.  Bezwada Wilson, an activist who has long campaigned for the eradication of manual scavenging, explains that most of the drainage and sewage systems across the country are not well planned and were built without proper engineering oversight. Any solution would need to take into consideration all the resulting differences in infrastructure, he says: “It can’t be that you come up with an alternative and force it upon the drainage system without understanding its nature.” Hamaad Habibullah is a freelance journalist based in New Delhi. 

When I picked up my daughter from summer camp, we settled in for an eight-hour drive through the Appalachian mountains, heading from North Carolina to her grandparents’ home in Kentucky. With little to no cell service for much of the drive, we enjoyed the rare opportunity to have a long, thoughtful conversation, uninterrupted by devices. The subject, naturally, turned to AI.  “No one my age wants AI. No one is excited about it,” she told me of her high-school-age peers. Why not? I asked. “Because,” she replied, “it seems like all the jobs we thought we wanted to do are going to go away.”  I was struck by her pessimism, which she told me was shared by friends from California to Georgia to New Hampshire. In an already fragile world, one increasingly beset by climate change and the breakdown of the international order, AI looms in the background, threatening young people’s ability to secure a prosperous future. It’s an understandable concern. Just a few days before our drive, OpenAI CEO Sam Altman was telling the US Federal Reserve’s board of governors that AI agents will leave entire job categories “just like totally, totally gone.” Anthropic CEO Dario Amodei told Axios he believes AI will wipe out half of all entry-level white-collar jobs in the next five years. Amazon CEO Andy Jassy said the company will eliminate jobs in favor of AI agents in the coming years. Shopify CEO Tobi Lütke told staff they had to prove that new roles couldn’t be done by AI before making a hire. And the view is not limited to tech. Jim Farley, the CEO of Ford, recently said he expects AI to replace half of all white-collar jobs in the US. 
These are no longer mere theoretical projections. There is already evidence that AI is affecting employment. Hiring of new grads is down, for example, in sectors like tech and finance. While that is not entirely due to AI, the technology is almost certainly playing a role.  For Gen Z, the issue is broader than employment. It also touches on another massive generational challenge: climate change. AI is computationally intensive and requires massive data centers. Huge complexes have already been built all across the country, from Virginia in the east to Nevada in the west. That buildout is only going to accelerate as companies race to be first to create superintelligence. Meta and OpenAI have announced plans for data centers that will require five gigawatts of power just for their ­computing—enough to power the entire state of Maine in the summertime. 
It’s very likely that utilities will turn to natural gas to power these facilities; some already have. That means more carbon dioxide emissions for an already warming world. Data centers also require vast amounts of water. There are communities right now that are literally running out of water because it’s being taken by nearby data centers, even as climate change makes that resource more scarce.  Proponents argue that AI will make the grid more efficient, that it will help us achieve technological breakthroughs leading to cleaner energy sources and, I don’t know, more butterflies and bumblebees? But xAI is belching CO2 into the Memphis skies from its methane-fueled generators right now. Google’s electricity demand and emissions are skyrocketing today.  Things would be different, my daughter told me, if it were obviously useful. But for much of her generation, she argued, it’s a looming threat with ample costs and no obvious utility: “It’s not good for research because it’s not highly accurate. You can’t use it for writing because it’s banned—and people get zeros on papers who haven’t even used it because of AI detectors. And it seems like it’s going to take all the good jobs. One teacher told us we’re all going to be janitors.”   It would be naïve to think we are going back to a world without AI. We’re not. And yet there are other urgent problems that we need to address to build security and prosperity for coming generations. This September/October issue is about our attempts to make the world more secure. From missiles. From asteroids. From the unknown. From threats both existential and trivial.  We’re also introducing three new columns in this issue, from some of our leading writers: The Algorithm, which covers AI; The Checkup, on biotech; and The Spark, on energy and climate. You’ll see these in future issues, and you can also subscribe online to get them in your inbox every week.  Stay safe out there. 

Earlier this year, the $800 million Vera Rubin Observatory commenced its decade-long quest to create an extremely detailed time-lapse movie of the universe. Rubin is capable of capturing many more stars than any other astronomical observatory ever built; it also sees many more satellites. Up to 40% of images captured by the observatory within its first 10 years of operation will be marred by their sunlight-reflecting streaks.  Meredith Rawls, a research scientist at the telescope’s flagship observation project, Vera Rubin’s Legacy Survey of Space and Time, is one of the experts tasked with protecting Rubin’s science mission from the satellite blight, which could make observations more difficult because the satellites are millions of times brighter than the faint stars and galaxies it hopes to study. Satellites could also confuse astronomers when the sudden brightening they cause gets mistaken for astronomical phenomena. An unexpected path When Rawls joined the Rubin project in 2016, she says, she had no clue what turn her career would take. “I was hired as a postdoc to help build a new imaging pipeline to process precursor images [and] analyze results to identify things we needed to fix or change,” she says. But in 2019, SpaceX began deploying its internet-beaming Starlink constellation, and the astronomical community started to sound alarm bells. The satellites were orbiting too low and reflected too much sunlight, leaving bright marks in telescope images. A year later, Rawls and a handful of her colleagues were the first to make a scientific assessment of the satellite streaks’ effect on astronomical observations, using images from the Víctor M. Blanco telescope (which, like Rubin, is in Chile). “We wanted to see how bright those streaks were and look at possible mitigation strategies,” Rawls says. Her team found that although the streaks weren’t overwhelmingly bright, they still risked affecting scientific observations.
Streak removal  Since those early observations, an entirely new subdiscipline of astronomical image processing has emerged, focusing on techniques to remove satellite light pollution from the data and designing observation protocols to prevent too-bright satellites from spoiling the views. Rawls has become one of the leading experts in the fast-evolving field, which is only set to grow in importance in the coming years. “We are fundamentally altering the night sky by launching a lot more stuff at an unsustainably increasing rate,” says Rawls, who is also an astronomy researcher at the University of Washington. 
To mitigate the damage, she and her colleagues designed algorithms that compare images of the same spot in the sky to detect unexpected changes and determine whether those could have been caused by passing satellites or natural phenomena like asteroids or stellar explosions. A rising force The number of satellites orbiting our planet has risen from a mere thousand some 15 years ago to more than 12,000 active satellites today. About 8,000 of those belong to SpaceX’s Starlink, but other ventures threaten to worsen the light-pollution problem in the coming years. US-based AST SpaceMobile, for example, is building a constellation of giant orbiting antenna arrays to beam 5G connectivity directly to users’ phones. The first five of these satellites—each over 60 square meters in size—are already in orbit and reflecting so much light that Rubin must adjust its observing schedule to avoid their paths.  “So far, what we’ve seen with the initial images is that it’s a nuisance but not a science-ending thing,” says Rawls. She remains optimistic that she and her colleagues can stay on top of the problem. Tereza Pultarova is a London-based science and technology journalist.

Overthink This is a podcast in which two very smart people (who happen to be young and hilarious professors of philosophy) draw unexpected philosophical connections between facets of modern life. Ellie Anderson and David Peña-Guzmán have done hour-long episodes on everything from mommy issues to animal justice, with particularly sharp segments on tech-adjacent issues like biohacking and the relationship between AI and art. Whenever I think society is dealing with a brand-new problem, these two unearth someone who was pondering it centuries ago. It’s a treat to listen to.  A film from the tech billionaire bunker Over the summer I was eager to watch Mountainhead, a darkly funny film by Jesse Armstrong, the creator of Succession, that follows four unlikable tech founders as they watch the world collapse under political turmoil and violence caused by AI deepfakes. I was prepared for it to seem like a documentary, but to a reporter who is in frequent dialogue with AI’s movers and shakers, it felt a little too real. From their remote mountain mansion, they talk about AI accelerationism, utilitarian ethics, uploading one’s consciousness to the cloud, liberating humanity to other planets—all common conversation topics among the tech elite that has had so much influence in the current administration.   Music by human beings For much of last winter I was reporting a story about just how far AI-generated music has come. As a lifelong musician (I play guitar, bass, and drums, none particularly well), I found the songs I heard—built with models whose creators have been sued for training on the discographies of artists without compensation—so convincingly human that they made me deeply uncomfortable. Since then, I’ve had a revitalized zeal for live shows where real people in punk bands or jazz trios do things that AI is not capable of (Sophie Truax is my latest favorite). 

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now A new technique from Zhejiang University and Alibaba Group gives large language model (LLM) agents a dynamic memory, making them more efficient and effective at complex tasks. The technique, called Memp, provides agents with a “procedural memory” that is continuously updated as they gain experience, much like how humans learn from practice. Memp creates a lifelong learning framework where agents don’t have to start from scratch for every new task. Instead, they become progressively better and more efficient as they encounter new situations in real-world environments, a key requirement for reliable enterprise automation. LLM agents hold promise for automating complex, multi-step business processes. In practice, though, these long-horizon tasks can be fragile. The researchers point out that unpredictable events like network glitches, user interface changes or shifting data schemas can derail the entire process. For current agents, this often means starting over every time, which can be time-consuming and costly. Meanwhile, many complex tasks, despite surface differences, share deep structural commonalities. Instead of relearning these patterns every time, an agent should be able to extract and reuse its experience from past successes and failures, the researchers point out. This requires a specific “procedural memory,” which in humans is the long-term memory responsible for skills like typing or riding a bike, that become automatic with practice. AI Scaling Hits Its Limits Power caps, rising token costs, and inference delays are reshaping enterprise AI. Join our exclusive salon to discover how top teams are: Secure your spot to stay ahead: https://bit.ly/4mwGngO Starting from scratch (top) vs using procedural memory (bottom) (source: arXiv) Current agent systems often lack this capability. Their procedural knowledge is typically hand-crafted by