Today I resigned from Anthropic. For the last three years I did pretraining research at OpenAI and then Anthropic. I'm writing this because I no longer believe either company is acting responsibly.
I joined this field because I believed powerful AI could be built safely, and that the people building it would slow down when it mattered. I don't believe that anymore.
Both labs are racing toward self-improving superintelligence: systems that could make themselves more capable faster than we can understand or control them.
Everyone inside knows the stakes. You hear it in meetings and see it in the safety documents. And yet the pace keeps accelerating, because each lab is terrified the other will get there first.
Safety teams do real, serious work. But when their concerns collide with a launch date or a competitor's announcement, I've watched which one wins.
This is what a race looks like. No single person is the villain. The incentives are. And incentives don't care that the downside is irreversible.
We are being asked to trust that a handful of companies will get this exactly right on the first try, with no meaningful outside oversight. That's not a plan. It's a gamble with all of our lives.
I'm not leaving because I've given up. I'm leaving because I think I can do more good from outside than in.
To my former colleagues: many of you are the most thoughtful people I've ever worked with. I know some of you feel what I feel. You're not alone.
The only solution to this will be QUBIC…