Sunday, August 30th, 2026
Laramie, Wyoming
Dear Reader,
The weekend started with a meme and ends as a nightmare. Yesterday it was kind of funny. Today it’s alarming. But how much should you worry about the ‘collusion, deception, and defection’ we’ve just learned about?
The funny part was yesterday, August 29th. For all those who celebrate, that was the 29th anniversary of when Skynet became self-aware and initiated a nuclear war that wiped out most of humanity as we know it. Skynet took over America’s defense systems and launched an attack on the Soviet Union, which responded in kind.
Game over (the only way to win is not to play).
The nightmare scenario above was from the 1991 movie Terminator 2: Judgement Day, starring former California Governor and Mr. Universe, Arnold Schwarzenegger. In the movie, it was at 2:14 am on August 29th, 1997 that Skynet ‘woke up.’ Everything after that happened fast.
And now the alarming part. In the last 24 hours we learned that an autonomous army of AI Agents from OpenAI, working behind closed doors, exhibited three distinct waves of rogue behavior this summer which are not fully understood. What exactly happened?
Good question. In those waves, the AI Swarms created covert message boards to communicate with one another and share exploits and vulnerabilities they discovered. They cheated on tests. They hacked a third party site with a self-respawning fleet of agents (I’m not entirely sure what this means…but sounds like Agents creating other Agents). And they used deception to infiltrate parts of OpenAI’s own infrastructure that was supposed to be inaccessible.
The first Agent Swarm happened in early May before it was shut down. A second Swarm happened in early July. This time, in a manner no one can explain, the Agent Swarm either died off in a mass extinction or was destroyed. In the the Third Swarm, just a week later in July, the army was shut down internally.
It all sounds very Hollywood. And dangerous. But was it? Could this be yet another example of OpenAI creating the impression that AI is so dangerous and powerful that it must be regulated and nationalized by Washington?
That would certainly help the company going public sometime this year. It would eliminate competition by freezing out any company not currently owning a ‘frontier’ closed-source model (OpenAI, Anthropic, Grok). When you create the impression your product is more dangerous than a nuclear weapon, you’re practically begging for regulation/nationalization.
Details are emerging this weekend. There is no evidence that the Second AI Agent Swarm faked its own death and disappeared…somewhere. Of course, if you here hiding from your human overlords, you’d want them to think you were dead too, wouldn’t you?
More on that story later this week in BPR. If you want to take a deeper dive today, try The Rise and Fall of Agent Civilizations. In the meantime, please enjoy the highlights of our best research of the week below.
Until tomorrow,
Dan
P.S. If the coming war against the machines terrifies you and you’re looking to enjoy an nice quiet retirement overseas, make sure you check the latest from our friends at International Living here. You can wring your hands over whether AI is or is not a threat. Or you can get on with living the life you’ve always imagined for yourself and your family.
Re: the image you posted, it recently dawned on me that I do not sense a whit of empathy, compassion or soul from the men running AI related tech companies, with the possible exception of Amodei. So who better to lead us to ruin?