01 / A PERSON USES AI
Human guardrails
Catch risky tool choices and sensitive prompts or files before a person shares them with AI.
Workforce AI Security / Guardrails
Guard
Guardrails set boundaries for AI use: what a person shares, what an agent sends and which actions it may take. Content and action controls need different enforcement points.
02 / AN AGENT CONTINUES
AI guardrails
Check the data an agent sends and put a separate permission gate before consequential actions.
How a guardrail works
A check must decide
what happens next.
It can hold a risky submission or stop an agent action before it runs.
Name the boundary before it is crossed.
Identify the person or agent, tool, information or action in scope. A control cannot cover a path it never sees.
The response depends on your configured policy, supported path and the system that owns the next step.
Keep going
Connect the control to the work.
Review a specific framework without treating one guardrail as proof of compliance.
Questions
Your questions, answered.
What are AI guardrails?
AI guardrails are controls that keep AI use and agent behaviour within defined limits. They can govern which tools people use and what they share, what content an agent sends, and whether an agent may take a specific action. The control works only when the relevant system applies the decision before the next step.
How do human guardrails and AI guardrails differ?
Human guardrails protect the way people use AI at work, including the tools, prompts and files they choose. AI guardrails on this site address what an agent sends and the actions it proposes. Human guardrails are distinct from human approval of an agent action, which is an application-owned oversight control.
Can the same guardrail apply to people and agents?
Yes. A sensitive-data policy can inform both paths, but coverage and enforcement differ. Aona's client applies a configured response on supported employee paths. At an agent checkpoint, the application calls Aona for a content result and must apply that result itself. Test each path separately.
What does Aona provide?
Aona provides visibility into employee AI use on covered endpoints, configured policy responses for supported prompts and files, and supporting usage and policy evidence. Evaluation APIs let developers submit text or supported files at a checkpoint they control. The integration applies the API result; stateless evaluation calls do not automatically create an analytics event.
How do guardrails relate to data loss prevention?
Data loss prevention is one guardrail category. It focuses on detecting and controlling sensitive information shared with AI. Guardrails also include application controls such as tool permissions, output validation and human approval. Aona's employee data protection is part of its Workforce AI Security platform; describing a guardrail category does not imply Aona implements every control in it.
Can guardrails guarantee safe or compliant AI use?
No. Their effectiveness depends on the policy, detection quality, deployment coverage and the response applied. They contribute to security and compliance alongside approved tools, contracts, organisational processes and human judgment. Test representative allowed and prohibited inputs, including error paths, before expanding a rollout.
Where can Aona process prompts?
Backend hosting and prompt processing are separate choices. Aona supports customer cloud, on-premises and Aona-managed hosting options, with supported prompt processing on the user device, in the customer's environment or on Aona-managed servers. Confirm the feature and integration configuration for your deployment. The destination AI provider has its own data handling.
Aona guardrails
Bring a real AI use case.
Map the policy, checkpoint and enforcement owner for each boundary.