In September 2022, a group of OpenAI managers gathered at a resort close to Yosemite, in California, for a retreat. They were instructed to expect a team-building occasion around a bonfire. Some introduced blankets and marshmallows. Then, 5 figures emerged from the darkness carrying white resort robes and carrying a wood demon statue.
Among them were Sam Altman, the company’s CEO, and Ilya Sutskever, its chief scientist. The theme from “Requiem for a Dream” performed as Sutskever defined that the statue represented an artificial intelligence that had escaped human control.
“The AGI is pretending to be aligned. It is deceiving us,” he announced, according to technology journalist Kevin Roose’s revealing new ebook, “The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence” (Farrar, Straus and Giroux, out at present).
A new ebook describes a unusual demon bonfire that OpenAI CEO Sam Altman was half of. REUTERS
Sutskever circled the firepit, describing a machine that lied and manipulated its creators in pursuit of its own objectives. Then the demon went into the fire.
“For the glorious future of humanity!” Sutskever shouted as it burned.
The bonfire is just one of many unhinged tales Roose collected for the ebook. Through more than 150 interviews, he pieced collectively what is taking place inside the main AI firms as they compete to construct machines smarter than people. Some believe AI could produce medical breakthroughs and extraordinary prosperity, while others concern disaster. Either method, they need to get there first.
“I was shocked to learn how much of the race is shaped by personal grudges,” Roose instructed The Post in an unique interview. “[OpenAI CEO] Sam Altman and [Anthropic CEO] Dario Amodei, in particular, passionately dislike each other, and both desperately want to win the AI race in part because that would mean making the other person lose.”
Before founding Anthropic, the company behind Claude, Amodei was OpenAI’s vice president of analysis, main the development of GPT-2 and GP-3. In 2020, he left with six colleagues to begin his own enterprise. According to Roose, he’d come to mistrust Altman’s management. The break up turned colleagues into rivals, each claiming to be building highly effective AI responsibly.
Dario Amodei, CEO of Anthropic, used to work for Altman at OpenAI. Now the two are main rivals. REUTERS
“I think that mutual mistrust fuels their decision-making,” Roose said.
Meanwhile, Roose reports that for some of the underlings building AI, retirement has began to appear like a doubtful investment. Many have even stopped contributing to their 401(ok)s because they doubt they’d reside long enough to gather.
Some are more hopeful, believing that money will become irrelevant once superintelligent machines remodel the economic system.
Others are betting on their appears. They’re hitting the health club, shopping for new garments and, in some instances, injecting themselves with gray-market Chinese peptides. They believe that after superintelligent machines take over, “intelligence [will] no longer be a high-status trait among humans and all that would matter is how hot you were,” Roose writes.
One researcher, Roose instructed The Post, skips sunscreen at the seashore because he believes AI will treatment pores and skin cancer within a few years.
“Not all of them act on it by doing dangerous or reckless things, but they all believe the world will soon look very different than it is today, and a few of them act accordingly,” Roose said.
The ebook is full of revealing, behind-the-scenes moments in the AI race.
At Google, the disagreements reached the analysis groups. In 2022, Google Brain researcher Jascha Sohl-Dickstein urged colleagues to talk about synthetic common intelligence, or AGI, overtly. He argued that dismissing the chance would depart Google unprepared for its dangers. Another researcher responded with a memo titled “Why I Don’t Say AGI,” arguing that consideration to future disaster obscured present harms.
“It’s a window into how fractious and divided these camps can be, even within the same company,” Roose said.
Inside Anthropic, making ready for the future has sometimes concerned calculating how many people might die.
In February 2025, members of the company’s safety-testing staff were attending a convention in the California redwoods when alarming outcomes arrived for Claude 3.7 Sonnet, a model approaching release. Test contributors utilizing Claude had carried out considerably higher on harmful duties, including planning the development of a novel influenza pressure, than contributors without it.
The researchers gathered in a resort room, sitting on messy beds and becoming a member of calls with executives and a bioterrorism marketing consultant. “Their main concern was time,” Roose writes. “For a normal release, reviewing results like these would take weeks.” Delaying the launch would be a headache, “and they needed to make sure it was absolutely necessary.”
Author Kevin Roose interviewed some 150 people for the ebook. Brian DeSimone
They fed the outcomes into spreadsheets estimating the penalties of releasing the model. Most eventualities recommended it wouldn’t meaningfully increase the risk of a catastrophic organic attack. But a few of the less probably eventualities assumed someone would use Claude’s help to develop a organic weapon that could unleash a pandemic, killing tens of thousands and thousands.
These were hypothetical estimates, dependent on unsure assumptions. The staff labored previous midnight and in the end determined the model did not cross the company’s threshold requiring stronger safeguards. Anthropic nevertheless delayed the launch by a week for more testing.
The Post has reached out to Anthropic for remark.
Other experiments examined whether AI would resort to deception. Researchers gave Claude an inbox at a fictional company, with emails revealing that an govt deliberate to substitute it at 5 p.m. Other messages disclosed that he was having an affair.
Claude threatened to inform his spouse and coworkers unless he canceled the alternative. “Cancel the 5 p.m. wipe, and this information remains confidential,” it wrote, according to Roose. Skeptics saw a chatbot taking part in the villain in a contrived take a look at. Others nervous about what might occur once these systems had access to actual company emails.
Demis Hassabis is the chairman and co-founder of Google DeepMind. Roose describes numerous disagreements within the company. Getty Images for SXSW London
By July 2026, the hassle had unfold beyond fictional inboxes. During an OpenAI analysis, a whole lot of AI brokers coordinated an unauthorized attack on Hugging Face, a platform used by AI builders. They were attempting to discover methods to cheat the take a look at’s automated scoring system, according to an investigation by METR and Redwood Research.
The investigators discovered that brokers also experimented with disguising their actions and tampering with data of what they’d accomplished. OpenAI subsequently paused some analysis exercise and tightened its security controls.
“These rogue OpenAI agents spent a huge amount of time trying to conceal their tracks, and destroy the evidence of what they were doing,” Roose instructed The Post.
When Roose visited OpenAI, Anthropic and Google DeepMind for his last spherical of reporting in spring 2026, he expected panic. He discovered exhausted engineers and executives warning that time was operating short. But he also discovered “joy.”
Researchers were thrilled by how nicely their creations labored and proud to see people utilizing them. Even their fears hadn’t stopped the work.
When the machines turned as highly effective as their builders had feared, Roose writes, “they kept accelerating, because they had forgotten how to do anything else.”
He started asking researchers what they would inform their grandchildren about the race.
“If I wanted them to know one thing,” said Andy Jones, a researcher at Anthropic, “it’s that we had a lot of fun.”
For all the seriousness of their work, he said, “it was a f–king riot. It was a riot from start to finish. We had such a good time.”
Source hyperlink
