Why it matters
Capability without visibility creates a widening control gap. We look for infrastructure that turns opaque model behavior into evidence: what the system can do, where it fails, how it responds to adversarial conditions and whether its behavior remains stable after deployment changes.
Assurance becomes infrastructure
For frontier and enterprise AI alike, a one-time evaluation is not enough. Models gain tools, data access, memory and new deployment contexts. The relevant market therefore extends from pre-deployment evaluation into continuous monitoring, adversarial testing and technical assurance.
- Capability evaluations and benchmarking
- Interpretability and behavior analysis
- Red teaming and adversarial testing
- Runtime monitoring and evidence capture
- Formal and semi-formal verification methods
What we invest in
We are interested in products that make safety evidence operationally useful: faster deployment decisions, stronger access controls, better incident response or clearer accountability for high-impact systems.