Jeremy Kahn’s Post

You’re absolutely right about bypassing guardrails ���.. but I’m not sure if it’s called bypassing guardrails or bypassing gated functions (scroll to the end of this post to find the easter egg 🥚) https://www.linkedin.com/posts/rohanpinto_claudecode-ugcPost-7444866784480038912-k6Xn/

These leaks will likely be less consequential than most currently anticipate. In real AI research there is very little that is actually "frontier" grade or truly "proprietary" as AI is built on centuries of well known math. Leaks may only marginally accelerate inevitable commoditization of modern AI tooling itself.

Thanks Jeremy Kahn for covering this. I think the Claude code leak will have profound consequences. I spent some time going through the leaked code last night and there are already some very interesting findings. With model capabilities now converging (and open-source models are rapidly catching up) , I’d say 50% of Claude code’s moat is in the harness layer : orchestration , guardrails , searching etc. That’s especially significant given Anthropic has historically kept its approach tightly closed and hasn’t open-sourced its core work.

Yeeesh. Tough week for Anthropic after a very very good few weeks for Anthropic. And on the day OpenAI raised the most money in the history of raising money… As the AI World Turns, These are the days of our lives…

Great and timely coverage as always. Interesting times ahead. Looking into the material right now so not too many AI filtered things to say, just that I really like the naming convention: Mythos and now Kairos are both great.

I really believe people have to understand the risks of fully agentic models like Claude Code…