[DigitalToday reporter Chi-gyu Hwang (황치규)] Palo Alto Networks is strengthening security integration with Google Cloud and supporting checks on the execution behavior of autonomous AI agents.

On Sept. 24 local time, SiliconANGLE reported that Palo Alto Networks is integrating Prisma AIRS runtime security with the Google Cloud Agent Gateway to monitor agent tool calls and connections to external systems.

Autonomous AI agents can call external tools, retain information and continue tasks without additional instructions from people. Palo Alto Networks views this as requiring security standards to go beyond checking prompts and responses and to examine what agents actually do.

Spencer Telman, a senior product manager at Palo Alto Networks, said agents should not be treated like chatbots. "Agents are autonomous applications that act on behalf of users or other agents, and this autonomy can lead to dangerous actions such as deleting a production database. Tools and memory can also be manipulated," he said.

Palo Alto Networks plans to expand the scope of inspection this fall across user inputs and outputs for the Gemini Enterprise Agent Platform.

MCP connections create two-way risks. Sensitive credentials can be sent outside through agent requests, and a compromised MCP server could send back malicious links, malware, or instructions hidden in tool descriptions. Palo Alto Networks said indirect prompt injection included in tool descriptions could lead to the leakage of agent conversation histories.

Palo Alto Networks is also expanding model governance. The company provides a model scanning service in an agent format through the Google Cloud Marketplace. The service checks model file licenses, publisher verification and suspicious elements before deployment according to policies set in Strata Cloud Manager.