AI agent identity and security means treating each AI agent like a new employee: giving it its own login, only the permissions its job needs, and someone who watches what it does. It is also called identity for AI agents, or agentic identity. An AI agent is software that is given tools and access to a company's systems, and acts on its own to reach a goal. No single product does all three parts yet. Identity and security companies are each adding a part.
An agent is one kind of non-human identity, an identity that belongs to software and not to a person. Others are service accounts, which programs use to log in, and API keys. These identities never join, move or leave the company the way an employee does. So nobody remembers to remove their access.
Why does an agent need its own identity? In practice, developers often run an agent on a shared administrator key, or on the login of the employee who set it up. Then the agent can do everything that key or that employee can do. And when something goes wrong, the logs show the employee's name, not the agent's.
The first part is the login. The agent gets its own identity in the company's identity provider, the system that checks who is logging in. Instead of a permanent password, the agent receives short-lived tokens, digital passes that expire after a few minutes. Each agent also has a named owner, a person who answers for it.
The second part is permissions. The rule is least privilege: the agent gets only the access it needs. Identity security posture management (ISPM) maps what each identity can actually do, and flags permissions wider than its job. For an all-powerful account, privileged access management (PAM) lends access for one recorded session only.
The third part is watching. Every action the agent takes is recorded in the SIEM, where rules raise an alert when the agent does something unusual. The security team can also switch the agent off at once.
Say a logistics company in Gurugram deploys an agent that answers suppliers' questions about unpaid invoices. The agent gets its own identity, owned by the manager of the accounts payable team. It may read invoices and purchase orders. It may release a payment of up to ₹50,000, and anything larger goes to an accounts executive for approval. Its token lasts 15 minutes. One night, a supplier's email contains hidden text that tells the agent to change the supplier's bank account number. The agent has no permission to edit bank details, so the change fails. The failed attempt appears in the SIEM as an alert.
Agents are built on an AI agent framework and sold as AI agents for enterprise work. Those that can pay, through agentic payments, need these controls most. The page on AI agent guardrails explains the wider idea.