By Ravi Shankar Bera
On September 8, a 27-year-old researcher named Jacob Coxon resigned from Anthropic after about three years of pretraining research at OpenAI and Anthropic. His post on X said: "Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives." CNBC reported it was viewed more than 70 million times. TIME reported his reasons as two conclusions: things are speeding up, and they are not under control.
Anthropic's own Evan Hubinger replied publicly that he thinks the chance of AI killing all humans is above 10% within the next decade, and that Anthropic does not yet have a plan to solve alignment for superintelligence.
I don't know if Coxon is right. Nobody does yet. What struck me is that the thing he is afraid of has a name, and the name comes from Tegmark's book. So this is the first in a short series on Life 3.0, and we start with the idea everything else hangs on.
Life, sorted by what it can redesign
Tegmark doesn't define life by biology. He defines it by what a living thing can change about itself, using two words most engineers already know: hardware and software.
Life 1.0 can't change either one during its lifetime. Both come from DNA and change only through evolution, over many generations. His example is a bacterium, which swims toward sugar using a simple fixed rule built into its body.
Life 2.0 is us. Our hardware is still evolved, but we can rewrite a lot of our software. We learn languages, trades and professions. We change our minds about what matters.
Life 3.0 designs both. Its body and its mind are things it can rebuild.
The detail I liked most is a small one. Tegmark points out that your synapses store roughly 100 terabytes of information, while your DNA holds about a gigabyte, barely a single movie download. That is why a newborn can't arrive speaking English. There's no room for it in the genome. Learning, not inheritance, is what made Life 2.0 so much more capable than Life 1.0.
He also gives the timeline. Life 1.0 arrived about four billion years ago. Life 2.0, humans, about a hundred thousand years ago. And many AI researchers, he says, think Life 3.0 may arrive this century, possibly in our lifetimes.
Why that matters right now
Here's the connection I couldn't shake.
Coxon's whole warning is about "self-improving superintelligence." CNBC explains the concept as recursive self-improvement: an AI system able to design and develop its own successor without human help. It adds that this is not yet possible, but that labs including Anthropic and OpenAI have themselves warned it would make it easier for people to lose control.
Strip the jargon off that and you get Tegmark's third stage. A system that can redesign its own mind. In 2017 it was a thought experiment in a book. In September 2026 it's the subject of an argument between people who work at the companies building the most advanced AI.
Ars Technica quotes Coxon saying the risk sits less in today's models and more in the prospect of self-improving systems that could "hack anything, revolutionize any field overnight, and acquire real power and resources." Whether you buy that or not, you can see it's a Life 3.0 argument.
A fair note on caution
I should say this plainly, because the topic attracts a lot of drama.
Warnings like this are not new. Ars points out that Geoffrey Hinton left Google in 2023 with a grave warning, and that Anthropic safety lead Mrinank Sharma resigned in February with a letter saying "the world is in peril." The sky hasn't fallen. That doesn't make any of them wrong, and it doesn't make any single one of them proof.
Tegmark's own tone in the book is closer to a physicist who thinks a good outcome is possible and wants us to steer toward it. That's the tone I want to keep in this series: serious, not panicked.
What comes next
Tegmark's real question isn't whether Life 3.0 is coming. It's what kind of future we want when it does, and who gets a say. The book walks through a long list of possible outcomes, from very good to very bad, and asks readers to decide where they stand. The next pieces in this series take those ideas one at a time: the intelligence explosion, the problem of making machines want what we want, and what it means to stay human.
For now, one question to sit with. If learning let us leap past biology, what do we want to happen when something can also rebuild its own body and mind?
I'd like to hear your answer.
Sources: Max Tegmark, Life 3.0: Being Human in the Age of Artificial Intelligence (Alfred A. Knopf, 2017). TIME, Sept 9 and Sept 15, 2026. CNBC, Sept 9, 2026. Ars Technica, Sept 2026.
Disclosure: This piece was researched and drafted with AI assistance.
Top comments (0)