If I point an AI agent at SemanticFed, can it reach data my access rules would refuse?

No, an AI agent cannot bypass your access rules.

In SemanticFed[1], AI agents operate under the exact same permission and authorization architecture as human users, with no separate or privileged path to data.

As detailed in Make sure AI is not a way around your access rules[2], SemanticFed prevents policy bypasses through several core architectural guarantees:

• Single Gateway Enforcement: Authorization occurs directly at the workspace API gateway. Every operation—whether running a query, reading a semantic model, or authoring an entity—is checked against the caller's verified identity, role, and workspace scope.
• Identical Policy Path for MCP: SemanticFed’s Model Context Protocol (MCP) server calls the same gateway. As described in Governed Context for AI Agents[3] and Let an AI agent run your federation over MCP[4], an agent only accesses documented metrics and sources permitted by the caller's role, and cannot resolve resources outside its granted workspace.
• Auditable Agent Sessions: Agent tool calls pass through the same permission evaluation and can be logged and audited, as illustrated in the MCP Agent Sessions and Governance[5] concept.

Concept illustration(opens in a new tab) Illustration of the concept, not a screenshot of the actual product — MCP Agent Sessions and Governance[5]

(Note: SemanticFed is a pre-launch product. While gateway authorization is code-verified, confirmation gates for destructive actions such as entity deletions are actively being hardened.)

To learn more about connecting agents securely, explore the MCP agent federation use case[4] or contact the SemanticFed team[6].