Trust

Security at Cleo Powered

Current security controls and clearly stated limitations for Cleo's pre-launch customer, API, safety, and telephony boundaries.

Reviewed Content owner: Cleo Powered

Security boundaries in the current service

Cleo separates its public website, customer control plane, machine-facing Agent Gateway, and private telephony gateway. Public callers cannot select internal prompts, providers, caller IDs, or telephony credentials. Cleo's security posture is still being developed alongside the pre-launch service.

Implemented safeguards

Scoped authentication

Customer records are workspace-scoped; Agent Project keys are revocable and their usable values are not stored.

Bounded requests

Call inputs reject unexpected fields and enforce destination, schema, list-size, and duration constraints.

Purpose authorization

Allowed calls receive short-lived authorization tied to the exact payload; unsafe or unclear calls fail closed.

Private telephony seam

The public gateway does not receive SIP, carrier, ARI, or private telephony credentials.

Idempotent creation

A required idempotency key prevents duplicate call creation and rejects conflicting retries.

Controlled deletion

Authenticated customers can delete terminal calls and tasks within their own workspace.

Security limitations and user duties

  • Pre-launch access may be restricted while controls are tested and hardened.
  • Keep API keys and account credentials out of source code, browser storage, prompts, and support messages.
  • Rotate or revoke an Agent Project key immediately if it may have been exposed.
  • Minimize sensitive context and review a result before using it for a consequential action.
  • Do not attempt to bypass call-purpose, destination, duration, rate, or provider restrictions.

Report a security issue

Send a concise report to hello@cleopowered.com with the affected surface, reproduction steps, and impact. Do not include live credentials, unnecessary personal data, or data belonging to another person. Avoid privacy violations, service disruption, social engineering, and access beyond what is necessary to demonstrate the issue. A formal vulnerability-reward program is not currently offered.