Validation is the next priority.
The project is seeking stronger technical scrutiny, independent reproduction and adversarial evaluation.
Novel Symbolic presents Sanctuary V3
Control and evidence around AI-proposed actions.
Sanctuary V3 separates model proposals from execution authority, applies a defined external control boundary and retains evidence of what happened.
The project
The model may remain variable. The control boundary should be testable.
Project updates