Stampede Platform

Safety & limits

Built to refuse — ambiguous identity, low confidence or a policy breach gets a clear no, not a guess.

What is it?

Heads up
The AI is designed to refuse
Ambiguous identity, low confidence, a policy-breaching prompt — each gets a clear, structured refusal, not a best-guess answer. Refusing is a first-class outcome. Every action lands in the audit log with the time and who did it.

The AI Engine is built to refuse. Ambiguous identity, low confidence or a policy-breaching prompt each return a clear, structured refusal with a reason you can act on, rather than a best-guess answer. It never sends comms without approval, never merges guest records on its own, and never crosses between businesses.

Every prompt, draft, approval and refusal lands in an append-only audit log with the time, the person and the model version — entries cannot be edited or deleted, so your compliance team can review a full week of activity without anyone’s help.

The AI will — the AI won’t

// WILL
  • Read your live profiles on every prompt.
  • Back every claim with a real detail on the profile.
  • Log every prompt, draft, approval and refusal.
  • Refuse clearly, with a reason you can act on.
  • Record the model version behind any value.
// WON’T
  • Send comms without your approval.
  • Merge guest records off its own guess.
  • Cross between businesses with prompts or data.
  • Make up a value to fill a gap — it refuses instead.
  • Be overridden by an “advanced mode” toggle.
Win
The audit log is append-only
Entries can’t be edited or deleted from the console. A correction writes a new entry that supersedes the old one. Your compliance team can review a full week of AI activity without needing anyone’s help.

FAQs

Contact sales

See Safety & limits running on your venues

Thirty minutes is enough. Book a demo or send a question — both routes land on the same team.