Hackworth

It kinda just puts them on par with their competitors in that regard. Microsoft had massive code leaks in 2020 and 2022, and Google had leaks in 2024 and 2025. Still, it’d be nice if this delayed the IPO.

I agree it’s a fascinating look under the cover. But I don’t think it’ll hurt Anthropic that much. Only Claude Code’s source leaked, not Claude itself. Most people use Claude through the web chat or mobile app. If you’re using Claude Code from the CLI, you’re using an API key, which is also true if you’re using Claude in Cursor or any other tool. Anthropic makes the same amount of money regardless of where that API’s being called from.

This leak does mean Anthropic’s lost some dominance in the code editor space, which will probably be good for the industry as a whole. We’re already seeing open source improvements made from people sorting through and learning from the leaked code. The only real loss is face. Their whole identity is safety, and this is the second leak this month. Oh, and some of their plans for upcoming releases.

The paper is more rigorous with language but can be a slog.

Anthropic has some similar findings, and they propose an architectural change (activation capping) that apparently helps keep the Assistant character away from dark traits (sometimes). But it hasn’t been implemented in any models, I assume because of the cost of scaling it up.