Coram AI raises $35m to scale autonomous physical security agents

Coram AI’s $35 million Series B is another data point in the quiet arms race to instrument the physical world with AI agents. While frontier labs chase bigger models, Coram is using that capability to turn thousands of mundane CCTV streams and access logs into a continuously queried, agent-driven security layer. At 1,500+ sites across schools, factories and churches, systems like Deep Investigation start to look like a narrow but very real slice of embodied intelligence at scale.([moneycontrol.com](https://www.moneycontrol.com/news/business/startup/coram-ai-raises-35-million-in-series-b-funding-led-by-ansa-capital-battery-ventures-13947418.html))

For the race to AGI, this round underscores how much capital is flowing into "operating systems" for specific verticals. Coram isn’t building a general-purpose model; it’s wrapping commodity models in an application stack tuned to physical security workflows—querying months of footage, correlating door events, and handing back human-readable incident reports. That’s the pattern we’re seeing across sectors: specialized agentic layers on top of frontier models, each with its own data advantage and feedback loops.

Strategically, this raises the bar for incumbents in video security who still sell dashboards and NVRs, not autonomy. If Coram can prove that AI agents materially reduce investigation time and risk, it forces everyone from camera OEMs to access-control vendors to either build or buy similar capabilities. The long game is clear: whoever owns the agent orchestration layer for the built environment owns the data and the defensibility in one of AI’s largest untapped markets.