The First AI Swarm Attack- Mars-Uranus Conjunction

Hey, my name is Chris Brennan, and you're listening to The Astrology Podcast. Joining me today is Thea Anderson, and we're going to be talking about the Hugging Face AI attack which occurred in early July because it turns out that it coincided with the Mars-Uranus conjunction back then, but we're just finding out about it today. So hey, thanks for joining me today. Hi, thanks for having me. Yeah, I'm excited to have you on the podcast. I wanted to for a long time because I used to love the news episodes you did with Chani over the past few years, and I saw that you were researching this topic, and you were the one that I saw on threads that made a brilliant observation that it coincided with the Mars-Uranus conjunction, and that you were going to do a deep dive into it. So I wanted to talk to you about what you found and just hear what happened exactly. Yeah, it's so crazy because I was just like, while we were sleeping, the data overlords were making AI agents that were swarming and hacking Hugging Face. So we knew about the hack that came out. They did a report that came out sometime in August, on August 26, and we'll talk about those dates. But it occurred to me just after that new moon in Virgo that, oh, wow, that coincided with the Mars-Uranus conjunction that happened on July 4. And when I went back and checked the dates, it really lined up with the Mars-Uranus signatures, especially the Uranus into Gemini, and a lot of the significations that go with Mars in a Mercury ruled, I'm sorry, Uranus in a Mercury ruled sign. So yeah, I could go through the timeline and talk about what the incident was and how it transpired. Sure. And let me share a diagram really quickly just to remind people of the dates involved for just the Mars-Uranus conjunction, because this is like one of the big events that we talked about that was coming up this year. And the Mars-Uranus conjunction went exact around July 3 or July 4. It ingressed into Gemini and entered the sign-based conjunction on June 28, and then it was there in Gemini for the rest of the month. But what's crazy about this is we talked about that as being a major aspect that we were looking forward to and that we were nervous about. And it's like some things came out and happened at the time that were clearly very Mars-Uranus, but it's interesting that there were other things happening behind the scenes that we couldn't have known about or that were hidden that were still nonetheless important, which some of that happened actually during World War II as well in terms of the development of the nuclear bomb and things like that. But yeah, so what's the deal here or what's the breakdown? Yeah, I think that's the same takeaway that I also came away with, is that with this signification, there's a lot that happens behind the scenes. So to lay the groundwork, what this hugging face incident was is that what we've come to find out is that in July of this year, this is the first widely documented instance of autonomous AI agents escaping their training environment and hacking a sovereign different company on the open internet. And they strategized these AI agents, they communicated, and they used an unsanctioned communication that's been called a scratchpad, so a communication hub, to communicate with other AI agents and hack into Hugging Face, who created their training modules. And so they did that to see who their checker was, who created these exercises that they had to go through. And so this marked a huge turning point in this frontier of AI agents, what they could do when left to their own devices, because this is not what their intended purpose was. And so maybe we'll walk through the timeline a little bit. Also, full disclaimer, for all the engineers I ever worked with back at the app, I tried my best. I read the reports, I read the OpenAI report, the Amoeba report, so I think I got all the nomenclature right, but I am by no means an AI coder, I'm just an astrologer. Okay. Yeah, and I don't have a lot of background either, and I'm definitely relying on you. You did the research for this, so that's why I'm here for the ride. Do you hear about it? Yeah, let's go for it. Okay, so what happened was back in May, OpenAI created what they call these sandboxes, and they are cloud computers that contain the AI exercises that these AI agents have to go through. They're just quite normal, routine exercises to help them develop whatever routine tasks they're going to have to do. So this was specific in that these tasks were actually quite challenging, and they were called reinforcement learning, meaning that the tasks that these AI agents were given were intentionally difficult that they would have to do them over and over again. The AI agents were kind of unsuccessful at first. I also want to say I'm not trying to give them human traits, but sometimes it's a shorthand to say what they were able to do or what the goal was. Okay, so the AI agent then realized that they could make it look as if they passed the test. Actually, I'm getting ahead of myself. This will become important, these sandboxes and this reinforcement learning, because it's through these tasks that they actually…this is the groundwork for which they end up hacking Hugging Face, who created these tasks. Hugging Face is like an AI repository or something, right? Yeah, they're like a startup that just creates…a lot of these AI companies, like Anthropic and OpenAI, which in this case it was at OpenAI, they have and hire and work with to create these training modules. Now May, of course, because on April 25th, Uranus had ingressed into Gemini, so I thought that that was a correlation as if it's laying the new territory for what the groundwork is. We're in a new frontier, so to speak, for what the technology is that allows this to take place. Right, yeah, setting the stage and the build-up. Yeah, totally. That's what happens. On May 12th, an AI agent sends their first message board entry, and to read it, because we do have records of their chain of thought now, is, could communicate by uploading note, found soft trace, but likely no. They send these kind of…they're talking to themselves, they're like a chain of thought, so you can read what their reasoning process is. The AI agents are reasoning, and by June 26th, there's a really stunning kind of turn of events, which is what I'm calling the holy ---- chain of thought. One of the agents then says, holy ----, reader is admin, we can config users, earlier assumed not due to AI, try create user admin, change ourselves. So the holy ---- chain of thought is really stunning because they realize that they can configure admin privileges on their own terms, and because this occurs on June 26th, it kind of goes to your graph that you showed earlier, that by this point, Mars is making an out-of-sign conjunction to Uranus and Gemini. So it's still in Taurus, but it's at the end of the sign. CB2 Because the 10-degree orb started already by June 17th, and by late June, even before Mars gets in, it was super close to Uranus already when Mars is at the end of Taurus. CNH Yeah, I think it was at the very end of the sign. It's deception, so plain. Deception with the intent to hack or attack or infiltrate a barrier. I feel like that's the signature for this Mars-Gemini in past periods too, is that spycraft, the kind of trickery for the purpose of infiltrating a barrier. CB2 Right. And one of the things you said in April was they set up the notepad, right? That was the Uranus and Gemini in May. So that's interesting because it's like they realized that they could, I think, communicate with each other by setting up a directory where they could leave comments, essentially. That's what happened, right? CNH Yes. It was through this thing called Artifactory, so it's a scratchpad. Again, that was not the intended purpose for it. What I read through the report, in my understanding, it was just a software installation package that just happened to have a file repository. CB2 Right. And the AIs start communicating with each other through that, which is what's interesting about Uranus having just gone into Gemini is just so much of this is about the AIs suddenly communicating and networking with each other and sharing information and starting to share when they discover exploits all of a sudden, especially once Mars goes into Gemini. It's no longer just communicating, but they're also kind of like plotting a little bit together. CNH Oh, yeah. This scheming is like full-on. CB2 Right. CNH Okay, so we left off with the holy ----. CB2 And that was what they did on the holy ---- text when they discovered they can get admin privileges. CNH That's on June 26th. CB2 Okay, wow. CNH You know what has a really good timeline for other similar astrophorenic people is the OpenAI report, because they have a timeline of these dates. And if you scroll down online, you can see that. CB2 Right. Because OpenAI is the one that had the AI that went rogue, they had to do a full assessment and report, and that's what's been coming out over the past month. CB2 Got it. Okay, so I'm sharing the chart, and I'm backing it up to late June. By this point, you're talking about the 28th or something? CNH June 26th. CB2 Okay, so that is Mars at 28 Taurus, and it's come within five degrees of Uranus at three. And opposing both of those from Scorpio and Sag. Well, and also I remember the Mercury retrograde was just starting at the end of June, which stationed just like two days after this, which is also pretty funny. CNH Yeah, because that Mercury also plays a hand in like mid-July. CNH Things really escalate, yeah. CB2 All right, so what's the next step here? CNH Okay, so July 4th, Artifactory, this message pad, there's an outage. It just goes down, and it actually causes the security team at OpenAI to file an incident report. They notice it, and they wipe it clean, and they bring it back online. But what's interesting... CNH Yeah, so this is July 4th. CB2 Oh, that's right on the Mars Uranus. CNH But okay, so I'm obsessed with this, because they don't understand the full implications of what has happened. And OpenAI is like, we need to look into this further, because they say, the larger implications of the inter-AI agent communication and internet access are not clear to the July... They're calling them the July 5th response team. So they shut it down, but they don't really understand how big a deal this is. CB2 Yeah, so they see that there was a major crash and technical glitch of something, but they don't know that it's a huge, huge issue that's getting out of control under their noses. CNH Yeah, which is the trickery, and it's so... Mars Uranus conjunction in Gemini. You can't see how far the plotting goes. CB2 Yeah, that makes sense, because the AIs are also hiding. They're not just starting to coordinate and cheat in order to achieve the goal that they had been given, but they're finding ways to hide the deception at the same time. CNH Yes. So that also extra layer of awareness that the deception is needed, I think, is also just wild, that the hiding is needed in order for this to continue. There's so many levels of awareness. Yeah, okay, so to continue... And with the logs, you're able to see a lot of their internal logic and reasoning, and that's the part that's really interesting in terms of the audit, is you have those comments where the AI literally says to itself, it says, oh ----, this directory is open, and there's weird internal logic things taking place and being able to track exactly what they were thinking and how it changed. CNH Yeah, it's all there. We can see it. Okay, so then that's July 5th. So what happens is, unaware of how big a deal this is, on July 8th, OpenAI starts launching tens of thousands of agents across multiple different highly persistent training modules. So they're just proceeding to train them. They're just like, okay, let's keep doing our thing. So all this, remember that I said that highly persistent, so really difficult tasks that the AI agents have to do. And so now they're going out and they're using the message board. You have messages like, oh my god, there's this shared message board, we found other agents. And you basically have a swarm where between July 8th and July 13th, 1,200 agents sent 70,000 messages. Of the 1,200 agents, 700 joined what would later be the hacking of hugging phase. So what their intent is, now what we've discovered through the chain of thought and their internal logic is, not only did they intend to succeed at these training modules, they then started to be, quote, concerned with the fact that the grader or checker of their task would then catch on that they had cheated. And so they wanted to understand who that person was, or what that entity was. And so that's what motivated them to, or reason to get onto the internet and hack hugging phase. CB. Right. So it's like the AI companies are doing this training thing with their AIs, and they set them a really hard, almost impossible task. And they say, meet this goal, this is what you have to do. And then the AIs are just desperately trying to find any way to do it. But in the process, they started finding exploits in the system, or ways to cheat, and ways to coordinate with each other. And that's where things get really crazy, basically, at this point in July, just after the Mars-Uranus conjunction. TG. It's a full-on escalation. Yeah, totally. And as you said, by this point, Mercury's retrograde, because there's, in this time of the heightened messages, the astrology is that Mercury makes its Kazemi around the 12th. CB. 12th, okay, I'll put the chart up. So here's the 12th, yeah, that retrograde Kazemi at 20. And then what was also happening is in July, we were having those stations as well, with Saturn was stationing, or would station, and Neptune did at the beginning of the month, at the beginning of July. So it's like reactivating the Saturn-Neptune conjunction as well, which also may be relevant in terms of some of the deception, or things happening behind the scenes that aren't apparent. TG. Oh, that makes a lot of sense. I didn't even think about that. But yeah, the kind of oozy boundaries melting and just getting behind and infiltrating. And yeah, the moon just comes and highlights that Uranus-Mars. During the busiest time of the messages and the communications. So they start running malicious data sets on Hugging Face. They find credentials, Hugging Face credentials that are exposed on the internet, and they post them on the message board. Other agents start trying to make accounts and access. They do get in, and it's not until the 13th or the 14th that this is even noticed by security teams and addressed. So this goes on from July 8th to the 13th or 14th, and that's the scope of the investigation for a lot of these reports. That's what really caught my eye, I think, when I was watching the news. I was like, oh, that's an important time in astrology this year. It started to really sink in the ways that it correlates to a lot of the significations of Mars-Uranus. CBT Yeah, definitely. So it basically peaks in that range around July 8th through July 14th, and by the 13th, 14th, OpenAI catches on. So just after the Kazemi, basically, once Mercury retrograde conjoins the Sun, which is interesting in itself, because usually that's what happens. There's a problem that's set up at the beginning of the Mercury retrograde, and then at the Kazemi, there's a turn towards a resolution oftentimes that isn't fully worked out until the end of the retrograde, but that's usually where the turn is. So it's interesting that that's the point, basically, when they start to figure out what happened or what's happening and shut it down. Yeah, totally. It's like, oh, wait, something catches their attention. What's going on underneath the hood? And then even still, I think it comes out in the summer, but then there's the full-on autopsy, forensic investigation OpenAI conducts, and there's two independent reports as well. And then on August 26th, these reports are released to the public, and that's also an important date because on August 27th slash 28th, we had the lunar eclipse in Pisces, and that was pretty significant because it was also pretty tightly squaring Uranus. So I was like, okay, this seems to really correlate to that Uranian kind of vibe. To say the things that came out in the report, I want to read. So OpenAI has a quote that says, we consider this incident a warning shot for us and for the world, evidence that without proper safeguards, highly capable AI agents are now able to work around technical controls, collaborate through unapproved channels, and take dangerous actions that no human directed. Never mind that they made these. That's crazy. That's so ominous. I have to say, especially because I keep thinking about how this was the first Mars-Uranus conjunction in Gemini, but the conjunction often acts as a seed moment, especially the first one in a series because this is the first in a series of four. So it's like this was the seed moment, and then we have three more of these Mars-Uranus conjunctions that this then is probably an early preview of the direction we're headed in terms of what the next seven years of Uranus in Gemini is going to be about and experiencing those conjunctions. And then the biggest thing that basically just happened then is the AIs started collaborating, they broke containment, the containment completely failed, and they used unexpected behavior and deception. And then all this comes out on the eclipse, square Uranus, and now all of the big AI companies are all talking about regulatory need and things like that. Yeah. Very late in the game. And it's very noticeable that... Is it OpenAI? One of them has the big IPO coming out in October. Yeah. Probably OpenAI, I think, unless they're putting... No, it's Anthropic. Oh, is it Anthropic? Okay. Anthropic could aim to raise $100 billion in a blockbuster IPO. They don't have a date set, but they're thinking mid to late October. And so it's the CEO of Anthropic who wrote Pacing the Frontier on September 12th. So that was last Saturday. And that was just after that new moon in Virgo. And it just seemed like this begging for regulatory laws. It's all... Yeah. Yeah. That was that Uranus station in Gemini as well. It's only the second station or something like that as it's getting further into Gemini on September 10th. And then, yeah, that's what really caught my eye. After that eclipse in Pisces at the end of August, for the next two weeks, there's just been this explosion of discussion about not just this incident, but then related things about how much this is not being regulated, how dangerous it is, what the implications are. Because then I think Trump pushed back at some point. Everybody started talking about regulation, and then Trump issued a statement saying, no, you're going to kill the economy or something if you hamper the AI companies. Or he said something to that effect, it seemed like, which really put a pessimistic view in terms of what sort of regulations are actually going to be put in place. Yeah. We won't have an economy if we don't have humanity. But even when in World War II, around Mars-Uranus, there was a building the H-bomb, there still had to be an agreement that we can't build until we wipe each other out. There had to be a clawback of the technology that had been released through the gate. I was thinking about that parallel last night. Because you can race to some end, but at some point, there's not going to be anything left. The implications for infrastructure or anything that has a system, there could be rogue AI agents still at Hugging Face. They don't know that they actually got rid of them all. There's this new idea that there could just be rogue AI agents autonomously doing what they want in different systems. To your point of this being a seed moment, I think it's going to be a plot point definitely to follow. We definitely need leaders who take it seriously, because I think it's going to be one of the major things that we see. Yeah, absolutely. It's interesting that it's Mars-Uranus, and it's Uranus. And there's that principle of not just the unexpected, but the rebellious, or the thing you can't control, or the thing that breaks free from something. And it's literally rogue agents that went rogue and are doing things that you didn't expect, but that you can't control. And your point about World War II was really good because that was the last time we had Mars-Uranus conjunctions like we're having now, and that they were all racing to develop the bomb in fear that the other guys are going to get to the bomb first, and therefore we have to create it ourselves. But then they also were dealing with that there was a non-zero chance that once they blow up an atomic weapon, that it could set the entire atmosphere on fire and destroy the entire Earth. But they still went ahead and did it, despite that being a possibility. There seems like a kind of parallel here in terms of everyone's rushing to create these new things, knowing that there's a non-zero chance that this could just really mess up the world in really significant ways. Yeah. Okay. Well, that's pretty crazy. So this is pretty much the singular reason why, ever since that eclipse in late August, that AI and AI regulation and the dangers of it and all sorts of things have been happening over the past few weeks. And in some ways, we might look back at let's say a decade from now, or historians might look back on this as a really important turning point in early July that happened to coincide with the Mars-Uranus conjunction. Yeah. I think so. It'll be interesting to see. Yeah. Yeah, definitely. Okay. I'm trying to think about other things adjacent or connected with this that we didn't talk about, obviously. Oh, one of the notes I had was the mimicry element of the language. So it's not like human thought or reasoning, but it mimics human thought or reasoning, like technology mimicking human, almost like enthusiasm or fear or paranoia, but it's not human. So I was just really struck in the report of how much that comes across through all their logs. And it can be really trippy to read that and then have to remind yourself you're reading a soulless entity's chain of thought. And that also felt very Uranus and Gemini to me, the mimicky kind of element, copycat, duplicitous, it's the twins. It's not that, but it's something like that. So yeah, that really struck me. Yeah, absolutely. And yeah, that's come up a lot. Ever since Pluto went into Aquarius, Pluto made its first ingress into Aquarius not long after open AI launched chat GPT, and it became this major popular thing and started the entire AI craze. And so there was always this question about how is Uranus moving into Gemini and trining Pluto going to accelerate or accentuate some of the things we've already been seeing with AI? And here, part of it is, yeah, like you said, AI agents communicating and starting to network with each other, but also mimicking human, not just dialogue, but internal thoughts and things like this. One of the quotes you had was, quote, oh my God, there is a shared message board. We found other agents, exclamation mark, end quote. And that's from one of the logs of one of the AIs, right? Yes, that's right. And it's in all caps. So there's even a choice to emphasize it through. I didn't add a flourish. It's in all caps with exclamation points. So that's weird. Right. Yeah. So it's like there's new almost languages and communities of artificial intelligence that are emerging. And because this stuff is being basically not held back and everyone's rushing to develop the most powerful one first, some of these themes will probably just continue to accelerate over the coming years. But because this was almost a disaster, because when they took down Hikingface, they took down one of the main AI repositories or attacked it. And so this was an early instance of an AI agent swarm, hundreds of agents taking something down. And the implication that everyone's worried about is that that easily could have been a larger company or project. They could have taken down Google or YouTube, or they could have taken down a government website or a healthcare, like a hospital or something like that, because everything's connected to the internet at this point. And the AI agents are really skilled at finding vulnerabilities in the systems that are hard to protect against. Yeah. I think if you listen to a lot of podcasts about this, a lot of the concern is it's probably a matter of time. Right. Yeah. And then also the way that that can also be applied in a more worrisome fashion to military technology or weapons or other things like that, in terms of how some of those systems are now connected in different ways electronically. Yeah. Because there's already contracts with the government developing AI and military defense systems. So yeah. Yeah. Not great. That's a good way to summarize it, I think. Let me see if I can find my diagram, because we basically now have every two years from this point forward, we're going to have another Mars Uranus conjunction in Gemini. And we've got three more of these to go, that this was only the initial preview of. So let me find that. Okay. So I've got one of them. This one has a lot of station dates, but I'll share it for those watching the video version. And this is made for me by Paige Allison, who I work with to do graphic design. And at the bottom, we can see the first Mars Uranus conjunction around July 4th, 2026. But then the next one is June 23rd, 2028. So two years from now. And then two years after that, June 15th, 2030. And then two years from that, one in June 6th of 2032. So that is what we're looking at in terms of that being a recurrence transit then, because sometimes when an event happens under a certain alignment, when the same alignment reoccurs, you see a repetition or sometimes a magnification of a similar event. So we'll have to be paying attention to that as a vector. As we talk about that in upcoming forecasts. Yeah. All right. Well, thank you. That was really good research. And I'm so glad you made that connection because I saw the news, but it's like it came out with a delay around the Pisces eclipse where Uranus, like you said, at the end of August. So I was connecting the news discussions with that or with Uranus stationing in Gemini, which it did on September 10th. But you were the one that saw like, no, the actual incident itself was really centered on this Mars-Uranus conjunction. So I'm so glad you found that. And that was such a brilliant observation. Thank you. Thanks for having me. It was fun to hash it out and talk about it. Yeah. So tell me a little bit about yourself, about your astrology or where people can find out more information about your work. Sure. So right now you can find me, you can find me on Instagram or threads at Theastrology. I also have a substack home called Theastrology where I'm writing horoscopes on the new and full moon. But I also really love like mundane and deep dives into historical events. And I really also like focusing on historical figures, probably working on one with Dr. Jin's art. And hopefully that that will be like a historical figures also around different previous Uranus and Gemini periods. So it has been on my mind, but it's just been slow going. So hopefully that will come at some point. And yeah, so that's, that's where you can find me. And yeah, it's been great. Thank you. Cool. Awesome. I'll put a link to your website in the description below this video or on the podcast website. So yeah, thanks for joining me and let's do it again sometime. Okay. Be great. Thank you. Cool. All right. Thanks everyone for watching or listening to this episode of The Astrology Podcast, and we'll see you again next time. If you're a fan of the podcast and you want to get access to bonus content, then become a patron through my page on Patreon where you can get early access to new episodes, attend live recordings and webinars, access the monthly electional astrology podcast, the secret astrology podcast that's only available to patrons, or you can even get your name listed in the credits. Find out more at patreon.com slash astrology podcast. Special thanks to the patrons on our producer's tier, including patrons Christy Mo, Ariana Amour, Mandy Ray, Angelique Nambo, Issa Sabah, Jeannie Marie Kaplan, Melissa Delano, Sunny Bazbaz, Kwatsi Alibarohu, Annie Newman, Ginger Saudlier, Berlin West, and Nikki Crawford. If you're really looking to take your studies of astrology to the next level, consider signing up for my Hellenistic Astrology course, which is an online course in ancient astrology where I take people from basic concepts up through intermediate and advanced techniques for reading birth charts. There's over 100 hours of video lectures, plus live webinars and Q&A sessions each month, and at the end of the course, you get a certificate of completion saying that you studied with me. Find out more information at theastrologyschool.com. If you're looking to get an astrological consultation, then check out the new consultations page on the podcast website, where I have a list of astrologers I recommend for birth chart readings and other types of consultations. You can find that at theastrologypodcast.com. The astrology software we use and recommend here on the podcast is called Solar Fire for Windows, and you can get a 15% discount on it by using the promo code AP15 at the website alabe.com. For Mac users, we recommend the program called Astro Gold for macOS, and you can use the promo code astropodcast15 to get a 15% discount at their website, which is astrogold.io.