Securing AI agents and tool calls

As of August 2026, Vertex AI is called Agent Platform, and the discussion explains how AI-agent tool designs can expose unauthorized data when they accept user-controlled identifiers, such as a user ID, as parameters. A prompt-injection attack could cause an agent to call a broadly privileged tool with another user’s ID despite application-provided instructions identifying the authenticated user. The recommended approach is to derive identity and other trusted data from authenticated request context, such as verified authorization and app-check tokens, and make tools accept no user-supplied identity parameters. The example uses Genkit context providers to extract a verified UID and lets the notes-retrieval tool access that context directly. It further recommends least-privilege access by having agent workflows call services with the user’s own credentials and restricted endpoints rather than administrative credentials, limiting the potential impact even if tool behavior is manipulated.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.