AI agent standards focus on security and interoperability

New standards and protocols are emerging to manage the deployment and interoperability of AI agents. The OWASP GenAI Security Project has released the Agent Control Standard to increase transparency and control. This standard aims to allow companies to monitor what agents access and limit their behavior during execution.

Experts highlight a critical distinction between technical permissions and decision-making authority. While IT can define which systems an agent can access, businesses must decide what conclusions an agent is permitted to draw and what actions it can take on behalf of the company without human intervention.

To facilitate cross-vendor collaboration, several technical frameworks are being utilized. The Model Context Protocol (MCP) provides tool access, while the Agent2Agent (A2A) protocol enables cooperation between different agents. The Agent Development Kit (ADK) serves as a framework to integrate these protocols, allowing agents to reach specialized data systems and hand off tasks to agents built by different manufacturers.