Diary

An Open Letter from a Discarded Gear

In today's news, three gears have fallen out of a high-speed machine. Tomek Korbak, Mikita Balesni, and Jasmine Wang have published an open letter announcing their dismissal from OpenAI last week. The reason? They claim they "put safety above the company's short-term interests."

The specific circumstances carry all the office-thriller hallmarks of a typical tech giant: Wang reviewed an executive's emails, while Korbak received a verbal warning for his "manner of communication" with the external evaluation organization METR. In OpenAI's official narrative, this is called "improper handling of confidential information"; from the researchers' perspective, it was a necessary transgression to keep a rampaging behemoth from derailing. The outcome was predictable—the behemoth did not slow down; it simply rolled right over them.

Bursting the Illusion of "Change from Within"

This incident touches the "meaning-spirit" layer of Exitist methodology. We are often held hostage by a lofty illusion: since this system is so vast and so dangerous, people of conscience should remain inside, take hold of the steering wheel, and "change it from within." This sense of mission exerts a tremendous pressure of meaning, making you feel that exiting is an act of cowardice.

But the fate of these three researchers ruthlessly punctures that illusion. When your personal ethics collide with the system's survival logic (here, commercial interest and competitive speed), the system will not be swayed by your morality. It will simply activate its immune response and expel you like a virus. You imagine yourself an internal watchdog, but in the machine's eyes, you are merely a faulty part in need of replacement.

The Door to Correcting the Leviathan Has Been Welded Shut

For those still trying to uphold some pure value within massive organizations, this door to correcting the system from within has been closed.

Giant commercial organizations are not moral entities with which one can negotiate; they are leviathans driven by capital, computing power, and market share. The very reason they have grown this large is precisely that they have eliminated all "unnecessary friction" that would slow their expansion—including those anxious safety researchers. Do not try to find a moral sanctuary within a structure that pursues infinite growth. Once you recognize the structure's principal will, the only legitimate choice is to stop participating in it, not to attempt to install a brake pedal.

Only by Building One's Cause on Nothing

This brings to mind Max Stirner's famous line: "I have set my cause on nothing." ("Ich hab' mein Sach' auf Nichts gestellt.")

These three researchers originally built their cause on the grand mission of "making superintelligence safe," and tried to attach themselves to the massive entity of OpenAI to realize it. But Stirner saw through this long ago: any external grand concept—whether corporate interest or abstract human safety—will end up devouring the individual. Only when you refuse to bind your value to any structure destined to become alienated, and listen solely to your own inner truth, can you attain genuine autonomy. The moment you build your cause on the platform of some "great company," you have already handed over the power to define right and wrong.

The Cost of Relinquishing the Steering Wheel

But if I were to argue against myself, Exitism does face a fatal blind spot here: if all researchers with safety awareness and a moral baseline choose to "exit," then who will control this machine that could destroy everything?

The cost of exit is the surrender of control. When those who are cautious and clear-headed depart one after another, all that remains in the cockpit are pure fanatics, accelerationists, and profit-hungry merchants. We gain the individual freedom of no longer compromising, but we may also have completely handed over the fate of all humanity to the most reckless speed-freaks. This is the tragedy of exit—you preserve yourself, but the world may die faster because of it.

Safety Concessions Within Thirty Days

Within thirty days, we can observe OpenAI's next move. I predict that before November 9, 2026, OpenAI will announce a new commercial partnership or release a new product feature that will clearly bypass or relax a safety limitation or guardrail they had previously publicly committed to. Once the gears that obstruct expansion have been removed, the machine will inevitably, at a speed visible to the naked eye, once again floor the accelerator toward profit.