NanoCo raises $12m seed for secure sandboxed AI agents

NanoCo sits right at the intersection of two hot trends: agentic AI and AI security. As enterprises experiment with autonomous agents that can read, write and act across infrastructure, the security model becomes critical. NanoClaw’s pitch is straightforward but powerful—run agents in hardened, containerised sandboxes that don’t get direct access to machine credentials or core services. The fact that the open‑source project went viral among both researchers and senior tech executives shows how much pent‑up demand there is for something safer than just giving an LLM your API keys.

In the AGI race, platforms like NanoClaw are important not because they push raw capabilities, but because they make it feasible to deploy increasingly capable agents without unacceptable risk. If we expect more powerful models and longer‑horizon agents, we also need containment and observability primitives that scale with them. The calibre of backers—Docker, Vercel, Monday.com, Slow Ventures—suggests this isn’t a niche security tool but a potential layer in the emerging “AI operating system.” Done well, it could slow down some of the riskiest failure modes (like unobserved lateral movement and data exfiltration) while still allowing aggressive experimentation with agentic workflows.