# AI Psychosis Is No Longer Fiction > Source: > Published: 2026-06-21 17:46:14+00:00 If you're into video games, especially those with deep, well-written stories, you've probably heard of Cyberpunk 2077. I played it last year, and I became completely hooked. Not just because of its stunning visuals or its complex mechanics, but because of the story. What amazed me most is how believable its world feels. Sure, it's fiction and it's set in the future, but in a strange way, it's also realistic, or rather future-realistic (I guess I just invented a new term). The trends we're seeing today in technology and the cyber world seem to be pushing us in the same direction as the world depicted in the game. And that's what makes its story so fascinating: it doesn't feel like pure fantasy. It feels like a glimpse of where we're headed. And it's quite terrifying, not gonna lie. There's a term in the game called Cyberpsychosis, a fictional dissociative mental disorder in the Cyberpunk universe. It occurs when an individual replaces too much of their organic body with cybernetic implants, gradually losing their humanity and the natural behaviors that define a human being. As their dependence on technology grows, empathy, emotional stability, and their connection to reality begin to deteriorate, often leading to unpredictable behavior. It's a fictional concept, of course, but it's also a fascinating metaphor for what can happen when technology starts to blur the line between human and machine. Even though Cyberpsychosis is an extreme, fictional scenario, the underlying idea doesn't have to be physical. Cognitive and psychological dependence on technology and AI can be just as significant, and in many ways, we're already there. AI has become an extension (and in some ways a replacement) of how people think, work, and make decisions. It's integrated so deeply into daily life that, for many, imagining life without it already feels unrealistic. And what makes this shift even more striking is how fast it happened, almost in a blink of an eye. The evolution of our minds and brains is much slower than the rate at which AI and technology are growing and developing. This creates a growing mismatch between human cognition and the environment we now live in, a kind of incompatibility that can potentially backfire over time. And in some ways, we're already beginning to see the effects of that gap in how we think, behave, and interact with information. People are strange about AI. They criticize it at every opportunity, yet many of them already rely on it daily. Whether they realize it or not, it has become embedded in how we write, search, create, and even think. If you look at social media, it's already clear how much content is AI-influenced or AI-assisted. But what many people miss is that AI isn't just another tool for reducing effort, it's a huge equalizer that changes what effort even means. If you simply use it to do the same things you were doing before, only faster, you're not really gaining anything. In fact, you risk falling behind. The real shift happens when you use it to expand what you're capable of producing with the same amount of energy. That's also why the fear of replacement feels so contradictory. People worry about AI taking their place, while at the same time actively using it in ways that make it easier for it to do so. In many cases, it's less about being replaced and more about gradually outsourcing parts of the work without redefining what their role should be. There's a parallel here with Cyberpunk 2077. Just like cybernetic enhancements in the game, AI offers immediate advantage. And just like in that world, people are often quick to adopt anything that gives them an edge, even when they're uneasy about the long-term consequences. People may express hostility toward AI in principle, but in practice, they're integrating it more deeply every day. And when access to it is suddenly removed or limited, everyone loses their mind. This tension is what I refer to as **AI psychosis**, not as a clinical condition, but as a cultural and psychological dependency pattern that's already starting to emerge. Let's talk about it in detail. On June 12, 2026, Anthropic released a statement regarding a U.S. government directive to suspend access to *Fable 5* and *Mythos 5*. If you skip the formal wording, the core of it is this: the U.S. government issued an export control directive requiring the suspension of access to both models for all foreign nationals. And the most striking part is this: even foreign national Anthropic employees wouldn't be allowed to use them. Everyone was excited about these models. There were posts, articles, and videos everywhere. In the AI space, it felt like the equivalent of a *GTA 6* release, hype at a global scale, with everyone watching closely. And then, almost overnight, it was gone. Every workflow that depended on those models was brought to a complete stop, unless companies or individuals had prepared a backup plan (which they should when working with models that just came out of the oven). Entire systems that had quietly integrated them suddenly found themselves disrupted, forced to adapt immediately or halt operations entirely. **So, where exactly did it all go south?** Now, everyone seems to have lost their minds over the ban. There are even more articles, posts, and videos circulating about it than before. In this article, I want to share my perspective and analyze the situation as a whole. I waited a week before writing it because I wanted everything to settle and the noise to fade a bit. When something has just happened, perceptions are often distorted and details can feel different in the moment. With time, the perspective usually shifts and that's exactly what I wanted before forming my thoughts. I'm going to objectively analyze the situation from multiple angles, including my own perspective. I believe I'm in a fairly neutral position to do that, since I'm not heavily invested in the *Fable* and *Mythos* ecosystem yet. I don't rely on AI models too deeply, I use them more as assistants than dependencies. Right now, I'm using Claude Sonnet (not even Opus), and it gets the job done perfectly well. Because of that, I don't fully relate to the reactions from AI influencers who go insane every time a new model drops. Of course, *Fable 5* is reportedly a major improvement over Sonnet and Opus, and it's not even close, but as much as I enjoy using AI for coding, I don't need it to always be at the absolute cutting edge. I'm comfortable with "good enough" tools. And it's interesting, because not long ago, models like Sonnet and Opus were considered state of the art, now they're already casually labeled as baseline/mediocre in some discussions. I also prefer a balance: I like using AI where it helps, but I still enjoy doing parts of the work myself. As stated by Anthropic itself, these vulnerabilities are not unique to *Fable* or *Mythos*, they can be found in other models as well. I find this particularly interesting, as it contrasts with some of the narratives that have surrounded these systems. I won't claim to know what on earth is going on behind the scenes, and I'm not interested in speculating. Instead, I prefer to focus on the information that is actually available and verifiable. There are plenty of conspiracy theories and rumors circulating around this topic, but I never trust rumors. I try to stick to facts and only facts, when forming an opinion. **Here's one key statement that should be taken into consideration:** "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers." So, it sounds like regardless of how many guardrails frontier models have, they still remain vulnerable in one way or another. This isn't exactly breaking news... it's more a reflection of how AI systems fundamentally work. Unlike deterministic systems, AI models are inherently non-deterministic. Even fully deterministic software can contain thousands of vulnerabilities, so it's not surprising that AI systems, especially those that behave in more probabilistic, human-like ways, also have weak points. In a sense, humans themselves are an example of this unpredictability: flexible, adaptable, but also inconsistent, error-prone, and sloppy when it comes to precise tasks. When you scale that kind of variability through AI systems operating at massive scale, the result can become pretty spooky. The jailbreak research was done in Amazon. Andy Jassy - the Amazon CEO himself was one of the tech leaders who raised concerns to the government of the United States. What's interesting here is that Amazon is one of Anthropic's biggest investors. At first glance, it doesn't make sense. However, when national security concerns are involved, priorities often shift in ways that override literally everything. In such contexts, security becomes the primary concern. The potential consequences of exposure or misuse are significant, and the last thing any system wants is to compromise sensitive information at a national level. This is especially important for a country like the US which is really on the very top in terms of AI and technological development, where the stakes are particularly high and the margin for error is extremely small. And when things get concerning at that scale, it's reasonable to expect that a country like the United States, that aims to maintain leadership across every major field, including the AI race, will take a close interest in it. History shows that even small mistakes can have significant consequences when the stakes are this high. The closest historical parallel is the space race decades ago, where technological and strategic dominance was treated as something that couldn't be compromised, even slightly. In races like that, maintaining an edge often means not giving up even an inch of ground to competitors. It's possible there was a miscommunication and that the ban was simply too harsh. I don't really know, time will tell. However, the way it was communicated by Anthropic themselves, gives the impression that this outcome may not have been entirely unintended. The framing around Mythos and Fable was, for some reason, based on fear, and it did create fear across the entire ecosystem. Whether that was the intention or just the result of cautious communication is hard to say, but the reaction it triggered was significant. Let's break that down a bit. Let's go back a bit, because this feels like a turning point in AI, where everything started, and where much of the later controversy can be traced. A new generation of models was introduced, with *Mythos* standing out as a major leap in capability, significantly more powerful, efficient, and effective than its predecessors. However, instead of releasing it publicly, it was kept closed. Rather than broad deployment, access was limited to a small number of major technology organizations, with the stated goal of securing and supporting some of the world's most critical software systems. Their cause with this new model was to improve cybersecurity for highly critical systems and help protect them from AI-powered tools and attacks. **See where's the problem here?** AI models have become so good at coding and scripting that, in some cases, humans can no longer reliably secure software against cyberattacks on their own. As a result, we end up building even more capable AI systems to defend against threats created by previous generations of AI. But then comes the next step: those defensive systems need to be even stronger to handle new, more advanced attacks caused by those new generations of models. And so the cycle continues, each layer of defense requiring a more powerful tool than the last. So where does it end? In practice, it may not. It's like inventing faster cars for the police to catch street racers. At first, it works. But over time, those same street racers get access to similar, or even better cars, and the cycle continues. Now you need even faster vehicles just to keep up. And the faster those cars become, the higher the stakes. When criminals gain access to more powerful tools, the potential damage they can cause increases as well compared to earlier, less capable systems. The same logic applies to AI. And it's not like this wasn't carefully considered, those people at Anthropic are pretty smart (well, duh). That's part of why access was limited to a small number of major companies: to strengthen the systems those organizations rely on, and to make them robust enough that a model like *Mythos* could eventually be deployed more safely at scale before releasing it to the public. **But there's a catch, and it's a big one:** If *Mythos* is as capable as it is claimed to be, it could potentially be used to improve itself over time, alongside the natural improvements made by Anthropic through updates and training. If that's the case, there is still a significant risk that such a system could be used maliciously against critical infrastructure, even if indirectly. The concern isn't just theoretical capability, but how that capability could be applied at scale. These systems can include, for example: And the list goes on. When you multiply that level of risk by the scale at which AI operates, the potential consequences can become absolutely catastrophic. **Here's what they wrote in the project announcement:** Zero-day vulnerabilities are something that were unnoticed by anyone. Nobody had any idea they existed the whole time... until Mythos found them. That's how good it actually is. **Alright how about this one:** Imagine discovering a vulnerability in a system known for its reliability, one that powers some of the most critical infrastructure in existence. Now imagine that vulnerability had been there for 27 years... Let that sink in for a moment. We're talking about something that dates back to the previous century, built and maintained by some of the most experienced engineers in the field. And yet, none of them were able to detect it, even with the advanced tools we have today. Tell me that's not insane. **Now this one:** And check out this tweet right here: So it really does act like a human. **But what's even more significant, is this:** As they mentioned, this is an operating system that runs the whole world. And that's not even an exaggeration. Now, that is actually pretty scary, because as beneficial these tools can be, it can also be a destructive weapon in the wrong hands. Especially today, when information and data is literally everything. On top of that, the risk doesn't only apply to systems themselves, it also extends to the people working within organizations. As mentioned, *Mythos* is described as having more human-like characteristics and behavior than previous models, which increases the likelihood that it could convincingly manipulate users into taking actions they shouldn't. This is particularly relevant in the context of phishing and scam attempts. Large companies are already primary targets for these kinds of attacks. If AI becomes advanced enough that it's nearly impossible to distinguish between a legitimate email and a malicious one, the problem becomes significantly more serious. There have already been numerous cases where companies suffered major damage simply because an employee clicked on a deceptive link or responded to a convincing message. Now imagine a scenario where those emails are generated by AI and appear completely legitimate, down to the smallest detail. **Oh, almost forgot, speaking of emails, here's what Sam Bowman, who's working as "AI alignment + LLMs" at Anthropic, Tweeted:** **If this tweet alone doesn't raise a petaquintjillion questions, I don't know what does.** But first of all, why on earth did that model even try to access the internet, let alone how it actually managed to access it? If it was restricted programmatically, well, it's less terrifying. However, if it was properly secured and isolated in the internal network, then it would raise some pretty uncomfortable questions about how it managed to reach beyond its intended environment... did it hack its way through the Anthropics internal environment? That's something straight out of a science fiction movie, if that's the case. Here are other tweets by him and Boris Cherny I don't know, this whole fear-based promotion is kind of strange. However I really do appreciate that they're completely open and honest about all that. This is something that should never be hidden. There are speculations and rumors circulating that Anthropic wants to be regulated by the government so they can potentially become the go-to organization for AI in the public sector. I have no idea whether that's true or not, I only heard it mentioned in a video by ThePrimeagen. There's also another rumor that this is being done for marketing purposes. But if that's the case, I think it may have backfired, because I don't believe banning both Mythos and Fable was their primary goal, unless they're playing a much, much longer game. The bottom line is this: if you say your AI is dangerous and terrifying, there's a pretty good chance people will believe you. So if it's all honest, the government will do everything it can to avoid compromising national security, and you shouldn't be surprised by that. Granted, there might be some miscommunication here and there, but how things appear often outweighs how they actually are. Sure, you can add as many guardrails as you want to *Mythos*, which effectively gives you *Fable*, but as mentioned above, guardrails can always break. And breaking the guardrails on *Fable* is much, much worse than breaking them on any other model, because when you break those guardrails, you essentially unwrap *Mythos* - the model that is capable of finding vulnerabilities in some of the most critical systems imaginable. On the other hand, if this is all just hype to attract more attention to the new models, then that's arguably even worse. Playing with something as sensitive as national security rarely ends well, and people will inevitably start asking questions. The whole situation is kind of weird, and we don't really know what's going on behind the scenes. But regardless, it all shows that AI psychosis is already a real thing, and we should do our best not to fall into that rabbit hole. AI can be an absolute blessing, I use it every single day, and it has made me at least ten times more productive, but we need to be smart about it. We need to find a balance between AI being fully controlled and AI being completely independent with no regulations. Neither extreme is good. If you're someone who jumps on the latest AI tools the moment they launch, take a moment and let things settle. Maybe give it a week or two until they're more thoroughly tested. And if you're running a company, always have a safety net, because AI is not deterministic, far from it. In fact, the more complex these systems become, the more unpredictable and potentially risky they are. Like every technology before it, AI is a tool. A very powerful one. But tools are most useful when they complement human judgment, not replace it.