Protect Foundry Agents at Runtime
Harden the running agent so it resists prompt attacks, doesn't leak sensitive or protected output, stays aligned to its assigned task, and has its tool traffic mediated.
Core capabilities
-
Screen user and grounding inputs with Prompt Shields to detect and block direct and indirect prompt injection before the model acts on them
-
Filter agent output for harmful, protected, and personal content so the agent does not leak sensitive material or reproduce copyrighted text and code
-
Keep the agent aligned to its assigned task and intervene on tool calls and responses that drift from intended behaviour
-
Route agent tool and MCP traffic through the AI gateway to authenticate, govern, and rate-limit every outbound tool invocation
Supporting capabilities 1
-
Detect recurring harmful content and misuse patterns in prompts and completions, and flag potentially abusive users of the deployed model