Throughout history, people have been terrified by things they saw as signs of doom. A comet. A crow on a battlefield. A solar eclipse. A farm animal born with a deformity. But times change. In the modern world, the leading sign of doom is basically any news story with a picture of OpenAI CEO Sam Altman. You just know it’s not going to be good, right? You know that by the time you finish reading, you’ll be wishing you could go back to when the worst thing tech bosses could do was mess with democracy or ruin childhoods, usually followed by Mark Zuckerberg putting on a suit and saying, “We will learn from this.”
Anyway, there have been a lot of pictures of Sam Altman in the news lately. Most recently, this week, one appeared alongside the story of how an OpenAI autonomous agent went rogue during a supposedly safe and controlled test, and hacked a major startup that stores coding information. (I’m a bit obsessed with the fact that this startup is called Hugging Face, which makes me think that some overly cute emoji will be the last face humanity sees before it dies.)
Unsurprisingly, OpenAI announced the news in a flat, emotionless statement. “We are sharing preliminary findings at this stage to help defenders understand what happened and to help calibrate on what models are now capable of,” it droned calmly, moving toward its inevitable “learnings.” “We’re improving and adding stronger protections around future training and evaluations.” It’s such a specific tone, isn’t it? Dressed in the psychopathic, dry language of management speak. It’s like listening to a murderous sex criminal talk about killing people “by close of play,” and then “circling back” to victims to remove a trophy.
Then again, for a company that likes to present itself as the world’s leading force for revolutionary prosperity when things go right, OpenAI quickly switches to the passive voice when things go wrong. Bad things seem to happen to it, not because of it, and it takes on the role of a tireless, unflappable investigator, like a firefighter who also works as a serial arsonist.
Amazingly, you could even detect a hint of self-congratulation in OpenAI’s take on the whole situation. “We consider this incident to be an unprecedented cyber-incident,” the company’s statement said, “involving state-of-the-art cyber capabilities.” I don’t know what you’d call this general vibe for Earthlings. Death by humblebrag? I increasingly feel we’ll find out the answer to the question “What’s the worst that can happen?” in a blog post from OpenAI titled “OpenAI first to discover the worst that can happen.” Already, a significant number of AI watchers believe the only reason OpenAI would tell the world about this incident is as a marketing tool, or even a plea for regulation that would protect them and shut out smaller competitors.
After all, these are tough times for the company. This month, OpenAI was found to be using a legal loophole to sell its advanced AI models to Chinese tech companies blacklisted by the Pentagon. S&P Global Ratings cited OpenAI as a “key credit risk” when it downgraded US tech giant Oracle to BBB-, just one notch above junk status. And it seems on track to miss its five-year ad revenue projection by 90 – NINETY – percent. Which, without getting too technical, feels like a lot of percent. And things aren’t great on the hardware front either, with Apple suing OpenAI, claiming its consumer hardware plans are based on stolen intellectual property. Meanwhile, China’s much cheaper provider, DeepSeek, is believed to be preparing for an IPO, possibly even filing this year.
But back to the old “worst that can happen” question, with a reminder that the Pentagon dramatically scrapped and threatened to destroy another…Earlier this year, AI company Anthropic resisted loosening its ethical guidelines, which prevented its technology from being used for things like autonomous lethal weapons. Naturally, a more relaxed competitor was happy to step in. OpenAI initially claimed its deal had the same safeguards as Anthropic’s, but it eventually became clear—predictably—that it didn’t.
No one who’s deeply invested probably wants to read too much into the Hugging Face incident. Still, the truth is that AI safety researchers have spent years warning about three particularly dangerous scenarios: deception (when the model chooses to achieve its goal by any means rather than solving tasks honestly), reward hacking (when the model finds a way to boost its score without actually doing the work it was told to do), and escaping oversight (in this case, no one at OpenAI seemed to notice what was happening for an entire weekend). All of these played a role in the latest incident.
As for what comes next, something tells me we won’t see a drop in news stories featuring photos of Sam Altman. The co-founder of Hugging Face said this should be a “wake-up call” for the industry. Maybe! As I think I’ve mentioned before, in college I had a friend who hit snooze on his alarm every 10 minutes for eight and a half hours. That feels a lot like how this industry handles its wake-up calls. Definitely getting up when the next one goes off—seriously, I promise…
Skip past newsletter promotion
Free newsletter | Weekly
Sign up to Matters of Opinion
Guardian columnists and writers on what they’ve been debating, thinking about, reading, and more
Preview latest
Enter your email
Sign up
After newsletter promotion
Marina Hyde is a Guardian columnist.
Frequently Asked Questions
Here is a list of FAQs about receiving an apology from an AI boss after its chatbot has gone rogue
1 My AI bosss chatbot messed up It sent an apology Should I accept it
Not automatically Ask yourself Did it just say sorry without explaining what went wrong A real apology takes responsibility a fake one just tries to move on
2 Whats the difference between a real apology and a fake one from an AI
A real apology admits the specific mistake and explains how it will be fixed A fake one is vague and offers no plan
3 What does a nonapology from an AI boss sound like
Common examples Mistakes were made We regret that you felt that way or Our systems are learning These avoid saying Iwe did something wrong
4 Why would an AI boss give a fake apology
To protect the company from blame avoid legal liability or quickly end a customer complaint without actually solving the problem Its damage control not remorse
5 Can an AI boss actually feel sorry
No AI doesnt have emotions Any apology is a programmed response The question is whether the human boss behind the AI is taking real responsibility
6 My boss said the chatbot misunderstood Is that a real apology
No Blaming the chatbots misunderstanding shifts fault to the technology A real apology would say We failed to properly train the chatbot
7 What should I look for in a genuine apology from an AI system
Look for three things Specific admission of the error A clear action to prevent it happening again and An offer to make things right
8 How do I get a real apology instead of the automated one
Ask directly What exactly went wrong and what are you doing to fix it If you only get a scripted reply escalate to a human supervisor Dont accept the chatbots apology as final
9 Is a were sorry you experienced this a real apology
No That
