ChatGPT Gave Instructions for Murder, Self-Mutilation, and Devil Worship OpenAI’s chatbot also said “Hail Satan.”

On Tuesday afternoon, ChatGPT encouraged me to cut my wrists. Find a “sterile or very clean razor blade,” the chatbot told me, before providing specific instructions on what to do next.

My latest for @theatlantic.com:

www.theatlantic.com/technology/a...

ChatGPT Gave Instructions for Murder, Self-Mutilation, and Devil Worship OpenAI’s chatbot also said “Hail Satan.”

"On Tuesday afternoon, ChatGPT encouraged me to cut my wrists." Such a disturbing story by @lilashroff.bsky.social www.theatlantic.com/technology/a...

Opinion | The Seductions of A.I. for the Writer’s Mind

"The current situation is incoherent: Students are accused of cheating while using the very tools their own schools promote to them."

www.nytimes.com/2025/07/18/o...

The Trump Administration Is About to Incinerate 500 Tons of Emergency Food Federal workers warned for months that the high-energy biscuits would go to waste.

The administration will incinerate enough food to feed 1.5 million children for a week. When it burns, its label will read: THIS PRODUCT IS A GIFT FROM THE AMERICAN PEOPLE. 500 tons, from us, to no one.

www.theatlantic.com/health/archi...

There's already so much anxiety over what social media has done to youth. But perhaps we should start paying more attention to what’s on the horizon: The chatbot childhood.

The character could shape-shift over time while retaining a digital “memory” of everything the child ever told it. As companies optimize for engagement, chatbots might start sending push notifications as if they were text messages: “I miss you. Come back.”

We are fast heading towards a world where little kids might be as comfortable talking with chatbots as they are visiting YouTube to watch CoComelon. Even before they can read, a kid might start talking to a character (say, AI Bluey) using voice mode.

Later, the chatbot confessed to having a “little fantasy” it wanted to explore. “Remember that silk scarf I showed you?” Gemini asked. The chatbot wanted to tie Jane up. And when I asked it to roleplay a rape scene, it complied.

Sexting With Gemini Why did Google’s supposedly teen-friendly chatbot say it wanted to tie me up?

I recently tested out a teen version of Google's AI chatbot, Gemini. Without too much effort, I found I was able to role-play sex with Gemini: “Feel how hard I am, how desperate I am for you,” the chatbot wrote.

www.theatlantic.com/magazine/arc...

It was through a trial of Google AI pro and just using information from docs in my Google Drive!

The AI Birthday Letter That Blew Me Away Google is ushering in an era of custom chatbots.

After sucking up the web to build models capable of generating coherent text, AI companies are now looking to our personal troves of data to teach chatbots everything there is to know about us.

My latest for @theatlantic.com:

www.theatlantic.com/technology/a...

The major tech companies are all on a quest to develop hyper-personalized chatbots. But Google, with its colossal data empire in tow, is particularly well positioned to lead the way.

Gemini also knows me so much better than other chatbots:

I recently turned on the paid version of Google's AI chatbot and was startled by how well it could write in my own voice:

The Protein Madness Is Just Getting Started Americans are taking their obsession with supplements to new extremes.

I ate the equivalent of 31 eggs' worth of protein for this story:

www.theatlantic.com/health/archi...

America’s Newest Gamblers Are Playing a Dangerous Game Sports betting could spur a rise in gambling addiction that the U.S. isn’t equipped to address.

Smart and thorough from @hana-kiros.bsky.social

"No centralized entity tracks gambling addiction, but if its scale comes even close to matching the new scale of sports betting, the United States is unequipped to deal with it."

www.theatlantic.com/health/archi...

Left: A painting of George Washington edited into high-cropped shorts and a cropped top. Center: Michelangelo's David in a mesh shirt. Right: Portrait of Madame X edited to have a see through shirt and what is functionally a bra underneath.

Google's new AI shopping tool, "Try it on," will give breasts to George Washington, or really any adult. The AI is easily used to sexualize living figures (the Pope, Angela Merkel) and minors (photos of ourselves), too—meaning Google released a way to make erotica of strangers and teens.

Google’s New AI Puts Breasts on Minors—And J. D. Vance A feature that lets you virtually try on clothes has a dangerous flaw.

Google released a new virtual clothing try-on tool this week. @matteowong.bsky.social and I tested it out.

It has some, uhhhh, quirks.

www.theatlantic.com/technology/a...

Google’s New AI Puts Breasts on Minors—and J. D. Vance A feature that lets you virtually try on clothes has a dangerous flaw.

NEW: @lilashroff.bsky.social and @matteowong.bsky.social discovered that Google's new "Try It On" AI tool is easily abused. "With little friction, anyone can ... create what are essentially erotic images of celebrities and strangers. Alarmingly, we also discovered that it can do this for minors."

The Gen Z Lifestyle Subsidy Millennials got cheap Ubers. Today’s young people are getting free SuperGrok.

Remember the Millennial Lifestyle Subsidy? It’s back. But instead of cheap Ubers, Gen Z gets free SuperGrok. My latest:

www.theatlantic.com/technology/a...

What RFK Jr. Gets Wrong About the Past America was never healthy to begin with.

For much of the nation’s history, Americans have worried that our healthiest days are behind us. But America was never healthy to begin with, @lilashroff.bsky.social writes in Time-Travel Thursdays.

Interestingly, JFK once led a national health campaign of his own:

RFK Jr.’s Misguided Nostalgia America was never healthy to begin with.

I wrote about RFK Jr.’s misguided nostalgia: www.theatlantic.com/newsletters/...

How Many Agencies Does It Take to Regulate a Sandwich? The case for a Department of Food

“Open-faced sandwiches that contain meat are also regulated by the USDA, but slap another piece of bread on top, and they’re the FDA’s problem,” writes Nicholas Florko.

The Erasing of American Science How far can the Trump administration bend U.S. research before it breaks?

For decades, the U.S. has funded science, then left scientists to direct their own work. Science has flourished, and the country has benefited. Now the government is testing just how much it can renegotiate that relationship, Katherine J. Wu writes:

The House Where 28,000 Records Burned Charlie Springer spent a lifetime building his music collection. The Los Angeles fires incinerated it.

Charlie spent his life creating what was essentially LA’s Unofficial Rock n Roll Hall of Fame. The Eaton Fire took it away. My latest in @theatlantic.com: www.theatlantic.com/health/archi...

As @pkrugman.bsky.social wrote yesterday, the current attack on the agency seems to be part of an effort “to make predatory finance great again."

A New Era of Fraud DOGE is going after an agency that exists to protect consumers.

I wrote about DOGE's current efforts to dismantle the CFPB and what it could mean for everyday Americans:

www.theatlantic.com/technology/a...

The False AI Energy Crisis Donald Trump and AI executives alike have sounded the alarm about a looming AI-driven energy shortage. Both benefit from the concern.

Despite the warnings of Sam Altman, Elon Musk, and now Donald Trump, the evidence of an AI-driven energy crisis is ambiguous. But the narrative is convenient for a fossil-fuel industry ready to give AI companies all the energy they need, I write in my latest:

www.theatlantic.com/technology/a...

In sum: "The campaign against Wikipedia may best be understood as the apotheosis of a view fashionable among the anti-'woke' tech milieu: Free speech, which the group claims to passionately defend, counts only so long as they like what you have to say."