The trust boundary is expanding
As AI systems become operational infrastructure, security cannot be reduced to cloud location or data residency. The system includes the model, credentials, connected tools, data pipelines, execution environment and control state.
Security needs verifiable control
- Protection of model weights and sensitive data
- Confidential or isolated execution where appropriate
- Attestation of software and runtime state
- Controlled access to tools and external systems
- Recovery mechanisms that preserve operator authority
Investment lens
We are interested in infrastructure that combines protection with control: products that reduce trust assumptions while preserving an operator’s ability to verify, revoke, migrate and recover.