Imagine a car manufacturer standing up at the premiere of a 2,000-horsepower supercar and saying, "Maybe we should calm down." That's exactly what Anthropic did with its Claude model. We're on the cusp of an era where artificial intelligence can improve its own engine. Is this the end of human engineering, or just a brilliant marketing ploy to make billions?
In the automotive world, we're used to raw power and clear specifications. When a new electric hypercar hits the market, we know what to expect. Basically, such a machine has 1,400 kW (1,877 hp) of power and a massive 2,360 Nm (1,740 lb-ft) of torque. Its top speed exceeds 412 km/h (256 mph), acceleration from 0 to 100 km/h (0 to 62 mph) and has you glued to your seat in a mere 1.85 seconds. With its electric architecture, it brings a battery whose battery capacity is 120 kWh, while the charging speed is rated at an incredible 500 kW at suitable fast chargers. All these are impressive numbers created by the human mind after years and years of testing. Engineers got their hands dirty to achieve perfection.
But imagine that same car being parked in a garage. At night, its computer wakes up. Without the presence of engineers, it rewrites the code of its inverter, optimizes the chemical structure in the battery cells, and adjusts the aerodynamics. You wake up in the morning and your car suddenly has twice the power and half the fuel consumption. Sounds like science fiction written by some overly enthusiastic Hollywood screenwriter, doesn't it? Well, welcome to the reality of the company. Anthropic, where this scenario is currently unfolding, except in the realm of programming code.
The fastest racer on the track calls for a red flag
Yesterday, June 4, 2026, the company that creates one of the world's most popular language models published a manifesto titled When AI builds itselfThe document, signed by head of internal research Marina Favaro and co-founder Jack Clark, was no ordinary press release about a new product. It was a call for a global, coordinated halt to the development of the most advanced AI models.
The reason? Recursive self-improvement. It's a term that inspires a mixture of absolute awe and quiet terror in engineering circles. It signifies the point at which a system becomes intelligent enough to independently design and build its own, much smarter successor. A machine that writes a better machine, which writes an even better machine. The human in this equation becomes just an observer with popcorn in hand.
Anthropic warns that this is no longer just a theoretical debate over a beer. The company's internal data is brutally clear. In May 2026, the model Claude wrote more than 80 percent of all new code the company integrated into its systems. The typical engineer combined eight times more code this year than in the same period in 2024. An eightfold increase in productivity in two years is not evolution, but a full-blown industrial revolution.
To understand the speed of this progress, let's look at the task-solving metrics. In March 2024, Claude was able to independently master tasks that took four minutes. A year later, he was able to do it for an hour and a half. Today, in 2026, he is paying attention and solving complex problems for twelve hours straight. The ultimate speed of this digital mind is currently immeasurable, and the acceleration in autonomous development is absolutely brutal. Their internal model, called Claude Mythos Preview, has achieved a fifty-two-fold speedup in improving machine code. It would take a human weeks of hard work to achieve a fourfold speedup.
"You want to be able to take your foot off the gas and step on the brake," Jack Clark explained to the BBC. In motoring jargon: when you're approaching a corner at a speed you don't even know if your tires can handle, it's wise to ease off the pedal a little.
A trillion dollars in the trunk
And here, dear readers, we must pause and raise a cynical eyebrow. History teaches us that when the largest corporations ask for regulation, they usually protect their market share. Anthropic is not just a scared group of academics. They have just overtaken OpenAI to become the most valuable artificial intelligence company on the planet. With a new round of funding, their value is approaching a whopping $1 trillion, and the company is preparing for an initial public offering (IPO).
So the question is: is the best racer on the track asking for a break because he is genuinely concerned about the safety of the audience in the stands, or simply because he is currently leading and wants the safety car to freeze his advantage over competitors like Google and OpenAI? Critics call this approach fear-based marketing. If you convince the world that your product is so incredibly powerful that it can destroy the world, you have just created the best possible advertisement for it.
However, the two are not necessarily mutually exclusive. The numbers they have published are real. The learning process of the next generation of models will require so much computing power that it is simply no longer controllable by existing social rules. Anthropic acknowledges that a unilateral shutdown makes no sense, as it would simply hand technological dominance over to China or less cautious competitors. They are calling for a global agreement, a kind of international treaty on the non-proliferation of digital weapons, with clear verification mechanisms.
Conclusion: Every machine has its bright spot
As a car enthusiast who has watched machines become faster, more complex, and sometimes dangerously powerful for decades, I have to admit that the whole situation fills me with a strange sense of optimism. Yes, the promise of a self-driving car is daunting. But let's look at it through the lens of everyday life.
Currently, for about 20 euros a month (the price of a good lunch), you can rent a version of Claude Pro. You get an assistant who doesn't sleep, doesn't drink coffee, and helps you analyze data at a speed that was reserved for supercomputers in government basements five years ago. The fact that the company behind this brilliant engineering feat is able to publicly and transparently point out the limits of its own technology is extremely refreshing in the world of arrogant Silicon Valley.
Perhaps much of this appeal is related to the upcoming IPO and building the image of a responsible leader. But in the automotive industry, the golden rule applies: the best drivers are not those who can just endlessly press the gas to the floor, but those who know exactly where the braking limits are and when to maintain control of the vehicle. Anthropic has proven that they may have the fastest engine in the world under the hood, but fortunately there are still people behind the wheel who understand the physics of collisions. And this is the most positive feature of this fascinating new model of the future.
