Arthur Conmy, who developed Automated Circuit Discovery while focused on mechanistic interpretability at Google DeepMind, has announced his move to Anthropic. His new work centers on aligning models during the training process itself, driven by the view that current systems like Claude still fall short for safely handing off AGI development tasks.
I'm joining Anthropic!
I'll start work on aligning upcoming models as they’re trained
Claude's capabilities are extraordinary. But like all models thus far, Claude isn’t aligned enough to safely delegate AGI development to
I can't think of a better place to work on this at
Conmy highlighted that even advanced models lack sufficient alignment for high-stakes delegation, positioning this hire as a direct response to that gap at a frontier lab.
No specific team, start date, or project ties have been confirmed, leaving open how this hire connects to Anthropic's existing alignment techniques.
Many users congratulated Arthur Conmy on joining Anthropic to align models during training because they viewed the work as impactful and worthwhile, while a few accused him of supporting a flawed company.
No Digg Deeper questions have been answered for this story yet.
I'm joining Anthropic!
I'll start work on aligning upcoming models as they’re trained
Claude's capabilities are extraordinary. But like all models thus far, Claude isn’t aligned enough to safely delegate AGI development to
I can't think of a better place to work on this at
Q: What do I mean by aligning upcoming models?
A: Triaging signs of misalignment in training, then aiming for root-cause fixes over whack-a-mole patches (e.g. training against the behavior). This post is the best work I’ve seen on how to align models: https://alignment.anthropic.com/2026/teaching-claude-why
I'm joining Anthropic!
I'll start work on aligning upcoming models as they’re trained
Claude's capabilities are extraordinary. But like all models thus far, Claude isn’t aligned enough to safely delegate AGI development to
I can't think of a better place to work on this at
I’ll move from London to SF for this work! 🥲
@ArthurConmy Congratulations ! Welcome to the bay ! hope you’re move goes smoothly!
@hopes_revenge @ArthurConmy Your*
@ArthurConmy Congrats Arthur, was fun working together!
@ArthurConmy congrats! see you in SF sometime soon 🙏
I'm joining Anthropic!
I'll start work on aligning upcoming models as they’re trained
Claude's capabilities are extraordinary. But like all models thus far, Claude isn’t aligned enough to safely delegate AGI development to
I can't think of a better place to work on this at
@ArthurConmy @tweem 500k views
@skooookum @ArthurConmy Your lucky im in a “charitable” mood
@ArthurConmy Congrats! super interesting (and impactful)
@Roprajo thanks Pratik! Likewise, sad it was brief!
@ArthurConmy Yay!! Congrats Arthur! 🥳
@ArthurConmy congrats arthur! 🙌🏻
@ArthurConmy godspeed
@ArthurConmy Congrats!!
I'm joining Anthropic!
I'll start work on aligning upcoming models as they’re trained
Claude's capabilities are extraordinary. But like all models thus far, Claude isn’t aligned enough to safely delegate AGI development to
I can't think of a better place to work on this at
@ArthurConmy Huge congrats!
@ArthurConmy Sad for Google but happy for you. Congrats!
@ArthurConmy yooo welcome!!
@ArthurConmy congrats!
@ArthurConmy Why is it google loosing another smart mind?
