The Download: what AI Observatory found on real AI use, plus Flock's camera trade-offs

MIT Technology Review's August 18, 2026 edition of its daily newsletter, The Download, opens with two feature items before running its usual curated list of tech headlines, a quote of the day, and a longer feature.
The first item, by Eileen Guo, covers a new research project called the AI Observatory. AI companies such as Anthropic and OpenAI regularly publish reports on how people use their products, but researchers say the companies only release the data they want the public to see, with no independent source to check it against. The AI Observatory sets out to fill that gap. Its analysis found many more sensitive personal behaviors in AI use than the usage reports published by major AI companies capture, since those reports focus more on work use than personal use. The project also found real differences between models: people were more likely to turn to Anthropic for coding, to Gemini for social and roleplay uses, and to ChatGPT for homework help.
The second item, by James O'Donnell, covers Flock Safety, described as the police tech giant behind a network of about 120,000 automatic license plate readers across the US. Flock recently announced platform changes meant to stop officers using the system for illegal or illegitimate purposes, including stalking. The piece pushes back on recent arguments defending Flock, such as the claim that the cameras can help catch a kidnapper or that they are otherwise harmless because nobody looks at the pictures. It argues those defenses skip the real question: what kind of crime-fighting system Flock chose to build. The network works the way it does because of specific decisions about what information to collect, who can search it, how long it is kept, and how widely it is shared, and those decisions set the actual terms of the bargain between security and civil liberties.
The newsletter's must-reads list of ten headlines starts with a child-privacy trial against Meta opening the same day: more than half of US states have joined the lawsuit, accusing Meta of deliberately designing addictive social networks and demanding changes including ending 'like' counts and infinite scroll, with four of the plaintiff states seeking $1.4 trillion in damages. Other headlines: Nvidia has committed up to $105 billion to OpenAI's planned Ohio data center, a site slated to cost up to $500 billion and come online in 2028, with OpenAI leasing the eight-gigawatt site for 20 years. Tesla is set to start Cybercab robotaxi rides in Austin this month, beginning with employee rides on public roads. Chinese robotics company Unitree has unveiled a humanoid robot it calls 'Superman' that it says can cover 12.66 meters in a second, faster than any human, ahead of the company's IPO on Wednesday. The remaining items cover an Amazon facility that scans and destroys rare books to train AI models, a report that Beijing is distributing AI training datasets reflecting 'mainstream Chinese values', an AI studio called Rogue building an AI video tool for uncensored adult content, a finding that women made up just 26% of new AI hires last year, AI research offering new clues about how breast cancer progresses, and an unearthed video suggesting Apple is adding cameras to AirPods.
The quote of the day comes from writer Carmen Hermosillo's 1994 essay on life online, quoted by the Guardian in a piece on warnings people ignored about the digital age: 'It is a black hole; it absorbs energy and personality and then re-presents it as spectacle.'
The longer 'One More Thing' feature, by Andrew Blum, covers Lincoln Electric System, a Nebraska utility. When a slow-moving blizzard hit the state, nearly 10% of Lincoln Electric's 150,000 customers lost power, and CEO Emeka Anyanwu watched the outage map as crews worked. The piece frames a bigger challenge facing utilities beyond any one storm: what the industry calls the 'trilemma' of balancing reliability, affordability and sustainability as electricity demand surges, driven partly by AI, while the industry tries to shift from fossil-fuel generation to renewables like solar and wind.
Key facts
- A new research project called the AI Observatory found people turn to Anthropic for coding, Gemini for social and roleplay uses, and ChatGPT for homework help, and found more sensitive personal AI use than company reports capture
- Flock Safety operates about 120,000 automatic license plate readers across the US and recently announced platform changes meant to stop officers misusing the system, including for stalking
- More than half of US states have joined a child-privacy lawsuit against Meta, with four of those states seeking $1.4 trillion in damages
- Nvidia has committed up to $105 billion to OpenAI's planned Ohio data center, an eight-gigawatt site slated to cost up to $500 billion, come online in 2028, and be leased by OpenAI for 20 years
- Unitree unveiled a humanoid robot called 'Superman' that it claims can cover 12.66 meters in a second, ahead of the company's IPO on Wednesday
Why it matters
The two lead stories both push back on convenient framings. The AI Observatory project disputes the idea that AI companies' own usage reports show the full picture, arguing sensitive personal use is undercounted because those reports focus on work use. The Flock Safety piece disputes the idea that a police camera network should be judged only by whether it occasionally helps catch a criminal, arguing the real stakes sit in the design choices, retention rules and access limits Flock made, and could have made differently. Set alongside a data center commitment that could reach $500 billion and a utility CEO watching a blizzard cut power to nearly 10% of his utility's customers, the edition traces how far AI's reach into money, infrastructure and everyday behavior has spread.
Who it affects
AI users whose chats turn out to be more sensitive than company usage reports disclose; police departments and the people whose plates are read by Flock's roughly 120,000 cameras; the more than half of US states now suing Meta over child-safety design, and Meta's users; OpenAI and Nvidia, tied together by the Ohio data center commitment; Unitree ahead of its IPO; and Lincoln Electric System's 150,000 Nebraska customers, exposed both to blizzards and to the industry-wide reliability-affordability-sustainability trilemma.
How to use it
Readers wanting the full AI Observatory methodology or the specifics of what a narrower Flock system could look like should follow the newsletter's links to the underlying MIT Technology Review features by Eileen Guo and James O'Donnell. The One More Thing piece by Andrew Blum goes deeper into Lincoln Electric System's plans. The must-reads list itself functions as pointers into other outlets' reporting, including CNBC, the New York Times, Reuters, Axios and the Guardian, rather than original Technology Review reporting, so readers chasing a specific figure should follow the linked outlet.
How solid is it
The AI Observatory and Flock items are original MIT Technology Review reporting with named bylines. The must-reads items are secondhand: each headline in the list attributes its figures to another outlet rather than to Technology Review's own reporting, so numbers such as the $1.4 trillion damages figure and Nvidia's $105 billion commitment rest on those outlets' reporting, not independent verification by the newsletter.
Risks and caveats
The newsletter does not say who funds or runs the AI Observatory project, nor give the full range of sensitive behaviors it found beyond the per-model breakdown. Flock's announcement date is given only as 'recently'. The $1.4 trillion damages figure covers just four of the states suing Meta, not the full coalition, whose total claim is not stated. Unitree's 12.66-meters-per-second speed claim for its 'Superman' robot is the company's own claim. Whether Rogue's uncensored AI video tool has actually launched or is still in development is also not stated.
“It is a black hole; it absorbs energy and personality and then re-presents it as spectacle.”
— Carmen Hermosillo, 1994 essay quoted by the Guardian