Robots Gone Wild
  June 26, 2026 Greetings, Meteor readers, I am obsessed with the story of this man who accidentally impaled himself and then just kept moving. I’m an avid hiker, but if there is so much as a pebble in my shoe, it sends me into a tailspin. Big ups to David and the friends who helped him get to safety.  In today’s newsletter, we are out of the woods and in the heart of the machine with podcast host and tech expert Bridget Todd, who tells us why we should care about an AI model going rogue. Plus, a little mood-boosting. Smoothing my hiking pole tips, Shannon Melero  WHAT’S GOING ON
Someone call Sarah Connor: Last week, as you’ve probably seen, an AI model created by OpenAI “escaped” its containment parameters and ran loose around the wilds of the internet for four days before it was discovered breaching another AI platform, Hugging Face. Now, if you’re anything like me, you’re probably thinking Girl, what the hell did you even just say? Which was precisely the first question I asked tech expert, author, and “There Are No Girls on the Internet” host Bridget Todd, who was gracious enough to explain why it matters that a robot has gone wild. “People should be alarmed,” she says. But let’s back up a tiny bit: What exactly happened? The TL;DR as explained by TechCrunch is that an internal model, created by OpenAI and not intended for public use, moved outside of OpenAI without being directly instructed to do so. It is, as far as the experts know, the first hack of its kind. “One point that many observers keep missing is that this wasn’t a rogue agent disobeying orders,” journalist Connie Loizas writes. “It was a system built to hunt for exploits, doing exactly that, just against the wrong target.” There’s another way in which we all might be missing the point about what happened, notes Todd. Much of the reporting “makes it seem like we’re in a Terminator situation where OpenAI’s models are so powerful that they’ve escaped containment and they’re coming for you.” That narrative allows OpenAI “to talk about their technology as if it’s something that they don’t have complete responsibility over.” But they absolutely do. The escape is a big deal “not because this is some super-scary, malicious sci-fi rogue AI, but because the humans who should be in charge and directing this are making decisions that lead to harmful outcomes,” she says.  WHAT THE HUGGING FACE BOTS HAVE NIGHTMARES ABOUT NOW (VIA GETTY IMAGES) In explaining this incident, OpenAI has employed the kind of language that places all of the blame on the model itself, as if it made a conscious choice to breach Hugging Face. But Todd says cybersecurity professionals have been warning about this kind of result for years—an AI model being powerful enough to exploit “known security vulnerabilities” in another system. Amid “all of the decision-making that OpenAI made as a company—how capable a model is, how strong the walls were that they built for it, what the goal that they set for it were, what safety roles they decided to switch off—there was no kind of law directing the duty of care and decision-making,” she says. So why isn’t there a law? If AI is so powerful and so omnipresent and we already have a mountain of data suggesting that unregulated AI is a danger, why aren’t politicians tripping over themselves to solve the problem before it gets any worse? “The United States really has a lot of anxiety that our country is losing the AI race with China,” says Todd. And the push to win that race, she believes, is taking precedence over common-sense policies. AI companies and government officials are essentially telegraphing: “We don’t give a fuck about y’all. Y’all are SOL,” Todd says. This sprint towards AI supremacy is part of why the technology has felt so inescapable lately. During our conversation, Todd and I swap examples of the often nonsensical commercials and billboards singing the praises of AI. Neither of us can remember the last time an industry tried this hard to ram something down consumers’ throats. “It’s almost like this kind of corporate marketing gaslighting speak,” Todd says. But if a product that is touted to be the greatest thing since the World Wide Web requires this much convincing, is it really all it’s cracked up to be? In the face of sky-high electric bills due to data centers, declining property values, and AI models changing the workforce, the powers that be are still wagering that the average American will just get on board because that’s what they’re being told to do. But “I can tell you that we will not,” Todd says. The breach of Hugging Face should serve as a wake-up call about the guardrails we need, so we don’t innovate ourselves into oblivion. It’s also a reminder that people are still in control; AI is not an inevitability we must all succumb to. Even the T-1000 was defeated in the end. AND:
 MOOD BOOSTEREvery Thursday, one of our editors shares a small thing that’s getting them through. If you’ve got something brightening up your week, reply to this email!  LAKE SOPHIA LOOKING DECEPTIVELY QUIET. (VIA INSTAGRAM) While solo-parenting this weekend, I impulsively decided to bring my baby and four-year-old to a new attraction in Kingston, NY, where we live. Nestled within Sojourner Truth State Park is a new spring-fed turquoise pool called Lake Sophia, named after one of the abolitionist and suffragist’s daughters. Despite that enticing description, the pool itself was not my mood-booster; the spot has been absolutely overrun since its opening earlier this month. By the time we walked 20 minutes from the overflow parking lot only to discover there were zero shady towel-spots left, I was ready to turn back. But then I saw my daughter’s bestie and her lovely mom and auntie, splashing in the water. They had a canopy tent! Snacks! Extra hands for the baby! And that’s what’s getting me through: communal caregiving, which saved my day. Neighborly co-parenting vibes 4eva. —Nona Willis Aronowitz  FOLLOW THE METEOR     Thank you for reading The Meteor! Got this from a friend? |
