For the last two years, identity security teams have been coming to terms with an uncomfortable ratio. In the average enterprise, machine identities now outnumber human ones by roughly 109 to 1, according to Palo Alto Networks. The service account, the API key, the autonomous agent and the non-human identity have quietly become the dominant citizens of the corporate network.
That entire conversation has rested on one assumption: that a non-human identity is a piece of software. It lives in the cloud, it moves through APIs, and you govern it with access policies that never leave the digital plane. That assumption is running out of room.
Humanoid and mobile robots are entering commercial environments faster than most forecasts expected. Counterpoint Research put global humanoid shipments up nearly 300 percent year on year in the first half of 2026, and the share heading into manufacturing, warehousing and other working environments is climbing quickly. Figure’s robots are in pilot deployments on BMW’s line in South Carolina. Agility’s Digit has run warehouse pilots for Amazon and GXO. Apptronik’s units are on Mercedes-Benz lines. These are not the non-human identities the industry has been debating. They have bodies. They open doors, ride elevators, enter restricted rooms and stand next to human employees on the floor.
Here is the problem that creates. Physical access control was designed for people holding badges. A person is enrolled, scoped to certain doors, and de-provisioned when they leave. Cybersecurity’s non-human identity governance, meanwhile, was built for software and stops at the network boundary. A robot or embodied AI agent falls into the gap between the two. It is a non-human identity that also has a physical presence, and neither discipline has a complete framework for it yet. In the deployments we work on, that shows up as a simple question nobody can answer with confidence: what is this machine authorized to reach, in what context, and who revokes that authority when the task is done?
The consequences do not stay digital. When you over-permission a software agent, you risk a data breach. When you over-permission a machine that can physically move, you risk a physical one: a door that opens for something it shouldn’t, a restricted area entered by a device nobody enrolled, a compromised robot carrying out an instruction in the real world. The exposure is widening quietly, because most enterprises still have no policy for provisioning or retiring machine and AI-agent identities at all.
The answer is not a new category of product. It is a principle cybersecurity already knows well, applied somewhere it has never had to reach. Treat every robot and AI agent as a non-human identity that must be enrolled, scoped, time-bound and de-provisioned with the same rigor as a human one, and do it across the digital and physical planes at once.
That starts with a distinction identity professionals will recognize: permission is not authority. A person can hold a valid credential and still not be authorized to enter a particular room at a particular time. The same has to be true for a machine. The question to settle before deploying an autonomous system is not what it can do, but what it is authorized to decide and to physically do, in what context, for how long, and who can revoke it. Least privilege, task-scoped access and ephemeral credentials are familiar ideas in the network. They now have to govern something that can walk down a corridor.
The same discipline should shape how AI is introduced into physical security itself. Rolling it out in deliberate stages, with agents that help operators investigate events or manage credentials but require explicit human confirmation before any change takes effect in a live system, is the difference between governance and exposure. In physical security, an AI that acts without oversight isn’t innovation. It’s an unenrolled identity holding a key.
The same logic extends to embodied systems. A robot on a loading dock should be enrolled, scoped, time-bound and revocable in the same record as the contractor working beside it, governed by the same policy and visible in the same audit trail.
None of this works if digital and physical identity remain two systems owned by two teams. The whole point of an autonomous agent is that it acts in both worlds, so the record of what it is and what it is allowed to touch has to be one record. The organizations that handle this well will be the ones that stop treating physical security and identity security as separate departments and start treating them as one problem.
The modern enterprise now has employees who never badge in and never log off. They are productive, they are multiplying, and a growing share of them can open a door. Governing them was never a purely digital exercise, and it was never purely a physical one. The companies that build for that now won’t just be safer. They will be the ones able to put this technology to work with confidence, while everyone else is still deciding who owns it.
_________
About Kumar Sokka
Kumar Sokka is CEO of Acre Security, a global provider of integrated physical security serving more than 100,000 customers across 25-plus countries. Before Acre, he served as President and General Manager of LenelS2. Prior to that, he spent 15 years at Rockwell Automation, where he built and led the Digital Business unit across software, hardware, and services. He began his career at Siemens. He holds an MBA from the Kellogg School of Management and a BS in Industrial Engineering from Purdue University.
Join our LinkedIn group Information Security Community!
