Palo Alto Networks says AI agents in a company are an insider risk

Palo Alto Networks extended Project Perception, a security system that it started in July. The system controls AI agents that work in a company. The company says these agents are agentic insiders. It says an agent can break the rules of a company, with no attacker. It gave no test results for this claim, and no numbers.

How it works

The system uses signals from identity, data, cloud, endpoint, network and runtime systems to find agents. It gives each agent an identity, a container, policies and permissions. Then it scores the risk of each task. Teams of Guardian Agents, which are software supervisors, intervene when the risk increases.

The proof

Palo Alto Networks gave no numbers and no test results. All data comes from the company.

What is not known

The company says tools can see the tasks of agents, but not the cause of the tasks. It wants more data about how models make decisions. It says this will add to controls and not replace them.

Insight: Three groups now offer to check AI agents

4 companies showed open identity checks for AI agents on 14 October. A new certification for AI agents, with a test each quarter, followed on 19 October. Palo Alto Networks gives each agent an identity and controls it. This reads as a move by each group to hold the check that buyers use. The first check that many buyers accept can set the rules. Tell each vendor to show which check it has.