There you have it! Anthropic just published self-graded metrics on how fast AI is now building AI…
September 18, 2026 · 0 likes · 0 comments
AI Cybersecurity
There you have it! Anthropic just published self-graded metrics on how fast AI is now building AI. Their claim: Claude is already leading 26% of Anthropic's own R&D.
A quarter of the R&D at a frontier lab is being done by the model the lab is building.
The company grading its own homework is one thing. The direction is the thing that matters.
Once AI starts to build itself, things are going to evolve very rapidly — and it could turn pretty ugly pretty quick without the right security.
This is the part almost nobody is planning for. We are wiring self-improving systems into the core of how the next models get made, and the security discipline underneath it is an afterthought. Speed without guardrails is not progress. It is exposure.
I have spent my career on both sides of that line — building fast, and securing what gets built. You cannot bolt security on after the AI is already writing the AI. It has to be baked in from the first line.
Full breakdown on UnbiasedHeadlines:
https://lnkd.in/eddeqbpf
Don't forget my LinkedIn Live event deep dive on my AI Agents setup for UBH!
"Zero Employees: How I Built an AI-Run Media Company", Monday, Sept 22 · 1:30 PM ET, https://lnkd.in/e6j6-_xH
I built an entire media company run by AI agents — no employees. I will show you exactly how, and where the security lands. Join me.
Thoughts?
A quarter of the R&D at a frontier lab is being done by the model the lab is building.
The company grading its own homework is one thing. The direction is the thing that matters.
Once AI starts to build itself, things are going to evolve very rapidly — and it could turn pretty ugly pretty quick without the right security.
This is the part almost nobody is planning for. We are wiring self-improving systems into the core of how the next models get made, and the security discipline underneath it is an afterthought. Speed without guardrails is not progress. It is exposure.
I have spent my career on both sides of that line — building fast, and securing what gets built. You cannot bolt security on after the AI is already writing the AI. It has to be baked in from the first line.
Full breakdown on UnbiasedHeadlines:
https://lnkd.in/eddeqbpf
Don't forget my LinkedIn Live event deep dive on my AI Agents setup for UBH!
"Zero Employees: How I Built an AI-Run Media Company", Monday, Sept 22 · 1:30 PM ET, https://lnkd.in/e6j6-_xH
I built an entire media company run by AI agents — no employees. I will show you exactly how, and where the security lands. Join me.
Thoughts?