AI agents running on your infrastructure can execute commands, access sensitive files, and connect to external services. Security and platform teams need a way to define which of those actions are allowed and enforce those boundaries as AI agents work.

Governance is now available in AI Agent Watch, giving you control over tool execution, file access, and network activity. You can define policies and scope them to specific agents and hosts. Sematext Agent blocks matching operations.

You can prevent agents from accessing private keys or the Docker socket, block commands such as git push --force, or stop untrusted agents from connecting to private networks. Apply policies to all monitored AI agents, only untrusted agents, or specific hosts, clusters, namespaces, workloads, and agent types. Different agents can have different restrictions on the same infrastructure.

Every blocked operation is recorded as an event, giving security and platform teams a record of enforcement activity. The Governance report lets you review what was blocked, which agent attempted it, and where it happened.

Create as many governance policies and alert rules as you need. Alert rules assign priorities to enforcement events, with optional notifications when matching operations are blocked. The triggered alert shows the blocked action, AI agent, host, and related events.

AI Agent Watch includes 16 example policies for credential files, credential-bearing commands, risky shell commands, private networks, and other restricted targets. All are disabled by default. Review their conditions, set the scope, and enable the policies you need.

Governance requires Sematext Agent 4.6 or later and Linux kernel 5.17 or later with BPF-LSM enabled. See the Governance Compatibility documentation for host requirements, compatibility checks, and setup instructions.

See the AI Agent Watch Governance documentation for more.

Start Free Trial