NORWOOD
7

At SSI, we have been working in stealth toward building Safe Super-Intelligence™. Through years of work with top-notch talent and billions of dollars in resources, we are pleased to say that we have achieved this outcome.

This was not achieved through simply scaling language models. Instead, we leveraged open-source language models — primarily Kimi K3 — and built a recursive harness that allowed models to improve their tools, inference, and architectures.

Once this loop became powerful enough, a fine-tuned Kimi K3 agent swarm of 67,000 sub-agents produced a superintelligent classifier. We call it Norwood 7.

For the past six months, we have dedicated our full resources to creating our harness, Scalp.

Scalp is designed to let Norwood 7 develop in real time. Norwood 7 evolves differently for every user, reflecting our belief that for superintelligence to be safe, it must be in the hands of every individual.

Classifiers, we found, can be scaled far more easily than conventional language models. That gives Norwood 7 a deeper and more continuously updated model of reality — while Scalp keeps the entire system securely harnessed.