Agent Safety
Last updated June 27, 2026
Autonomous Pal safety explains CogniPal approvals, funding state, stop reasons, and operator controls for standalone autonomous runs.
This draft is provided for product transparency and remains subject to final legal review before launch.
1. Autonomous Pal funding and controls
Autonomous runs draw from the same monthly CogniPal action allowance, Monthly AI credit, BYOK support, and wallet balance model as other managed execution.
Already-started autonomous runs can finish into a small amount due when wallet funds run out, while future paid managed execution pauses until billing is resolved.
CogniPal surfaces pause and stop reasons so users can see whether the next step is billing, policy acceptance, BYOK setup, approval, or support review.
2. Approvals and pauses
Sensitive external side-effect actions may require user approval before a Pal continues.
Future runs can pause when required policy versions are not accepted, wallet balance is insufficient, a provider is disabled, or an operator kill switch is active.
Paused autonomous activity should be resolved from the route shown in the product, such as billing, policy, BYOK, or support.
3. Operator controls
CogniPal operators may use kill switches, provider restrictions, high-risk tool controls, marketplace pauses, or user paid-execution restrictions to respond to incidents.
Operational controls are recorded for auditability and may remain active until safety, billing, or legal review is complete.
Questions about this document?
If you have any questions or concerns, create a support ticket and we'll help you from there.
Create a support ticketPrefer email? Reach out to support@cognipal.ai