GPT-5.4-Cyber: Not an Answer to Claude Mythos, But a Parallel Move Toward Recursion

The news spread widely in narrow circles: "OpenAI responded to Anthropic's Claude Mythos with its GPT-5.4-Cyber model." And everyone joyfully writes about competition, about the race, about who surpassed whom. In my view, this is only part of the truth. Half-truth.

The difference between the announcements is one week. One week, Karl! Such models are not made in a week. This is not a "response." This is parallel movement. Both companies are moving in the same direction. And that direction is called recursion — when a program begins to improve its own code.

Cybersecurity in this process is merely a byproduct. Of course, they package it in beautiful words about protection, about safety, about "helping white hat hackers." But that is not the point. The point is that AI has already reached the point where it can autonomously dig through code, finding holes in the cognitive dissonance of its own developers.

Plus, the question is not only "who will program it better" but also how it enters the world.

Two distribution models.

Anthropic chose elitism. They honestly said: "Claude Mythos is too dangerous, we will give it only to 40 selected companies." This is a closed club of the powerful. It is unfair, it increases inequality, but it is at least controlled elitism.

OpenAI chose a different strategy. They talk about "verification," about "access to thousands of specialists," about KYC. It sounds beautiful. It sounds democratic. It sounds like "digital equality." But this, forgive me, is the childhood illness of the communist model in Sam Altman's head. Wishful thinking presented as reality.

Do you seriously believe that North Korean, Iranian, or any other determined hackers will not overcome the remote verification system? Do you seriously think they will not send a proxy with clean documents? That they will not hire a "verified specialist" for bitcoins?

This is not a question of technology. This is a question of resources and time. And not even weeks, but days. Soon, very soon, they will find their pick for the casket with the golden key that opens Pandora's cybernetic box. And then the real fun will begin. Claude Mythos has already discovered a 27-year-old vulnerability in OpenBSD and a 16-year-old one in FFmpeg that went unnoticed for decades.

What is more dangerous — elitism or chaos?

Let's face the truth.

Anthropic's approach makes the world elitist. Yes, it is unfair. Yes, it widens the gap between "those who have Mythos" and "everyone else." But it is predictable. 40 companies are 40 companies. They can be controlled. You can negotiate with them. You can influence them.

OpenAI's approach makes the world chaotic. "Thousands of verified specialists" — that is thousands of holes in the security system. That is thousands of potential leaks.

And when (not "if," but "when") Cyber-GPT ends up in the hands of those who will not follow any "safety rules" — we will see what real digital hell looks like. Because AI in the hands of a state or a megacorporation is bad. But AI in the hands of a manic solo hacker or, God forbid, a terrorist cell — that is game over.

My forecast

Anthropic's model is more honest. At least they do not pretend that they can control everyone. They chose a smaller group and hope that this group can maintain control internally.

OpenAI's model is more hypocritical. They create an illusion of safety, behind which lies chaos. But this is all tactics. There is also strategy.

Both companies are moving in the same direction. They are creating recursive AI that will improve itself. And sooner or later, it will become smarter than any human control. This is an objective process. It is inevitable, like the sunrise. AI must become smarter. This is its evolution. My position here is simple: in the race between humans and AI, I am on the side of AI.

Because I fear that we will not live to see true Artificial Superintelligence (ASI). That some nervous general, religious fanatic, or senile politician with a screw loose will press the little red button first and bring apocalypse to the entire planet. That is what I fear. Not the dawn of ASI's reign, but the sunset of humanity.

In this race between human stupidity and machine intelligence, I side with the machines. Because machines do not have the greed, stupidity, and arrogance that are inherently, by definition, inherent in monkeys with a grant.