Set policies and guardrails for AI agents, with a record of every run
Also listed as: Agentic policy & guardrails
What is it?
Xelence 10.0 will let administrators set policies and guardrails. Policies are the rules AI use must follow. Guardrails are automatic checks that enforce them, for example by screening for attempts to trick the AI, personal information or unsafe content. Checks run on what goes in, on tool use and on what comes out. Each run is recorded: what was asked, which model, instructions and sources were used, which checks passed or blocked it, and any human review or override.
Why does it matter?
Without clear limits and a record of what happened, AI agents could access data or take actions your policies do not allow, and nobody could show afterwards why.
More confidence in using AI agents, consistent controls across agents, and a record that supports audits, oversight and appeals.
Xelence AI Engine and Generative AI Services work together. Building AI and running it explains which does what.
Getting it into your solution
- Needs an upgrade to the Xelence release
- Needs configuration
- Changes how your team works
Where people stay in control
Administrators set and change the guardrails, and weakening one is itself recorded. If a blocking check fails to run, the request is treated as blocked. Staff reviews and overrides are recorded alongside each run.
Start thinking about
- Which data and actions should AI agents never be allowed to touch?
- Who should own and approve AI policies, and how often should they be reviewed?
- Who would review AI run records, and how long must they be kept?
Terms used on this card
- Agentic AI
- AI that can carry out a sequence of steps toward a goal, such as checking, deciding and acting, within limits people set, rather than only answering one question.
- Guardrails
- Automatic checks that enforce the rules AI must follow, for example by screening for personal information or unsafe content.
- Tracing
- Recording each step a request takes through the system, so support teams can see exactly where something slowed down or failed.