trust-mode

AxonTrustMode

Make  AI  automation  stop  for  confirmation  before  risky  actions

A topic hub for teams that need controlled automation through confirmation cards, permission boundaries, accepted artifacts, and run evidence.

What Trust Mode Is

Axon Trust Mode is the confirmation and permission boundary inside an Agent run. It lets an AI digital employee continue low-risk work, but stop before external email sending, deletion, permission changes, form submission, customer commitments, or sensitive content handling.

Trust Mode is not paperwork added to automation. It makes critical actions understandable. A confirmation card should explain the action, source, impact, risk, available options, and accepted artifact so a human owner can decide from evidence instead of a vague popup.

Axon's Trust Boundary

Axon places Trust Mode inside the Assistant -> Run -> Build path. Assistant detects risk, Run presents confirmation cards and artifact evidence, and Build stores which actions require confirmation, which exceptions must stop, and which permissions the Agent may use.

System capability and skills can make automation faster, but they should not bypass the boundary. A reliable Agent knows when to continue, when to stop, who must confirm, and what record is created after confirmation.

Core Workflow

Start by listing high-risk actions in the task: send, delete, overwrite files, change permissions, submit externally, or make financial and customer commitments. Then define the confirmation card fields for each action, including source evidence, suggested action, impact scope, expiry, and rejection handling.

During Run, the Agent continues low-risk steps and produces a confirmation card at the risky step. After the owner accepts, rejects, or asks for more evidence, Axon records the result in the run history and accepted artifact chain for later review.

Common Use Cases

An email summary Agent can read the inbox, draft replies, and organize attachments, but it should stop before sending, deleting, promising delivery, or handling sensitive content. Browser form automation can fill fields, but the final submission should show source, target page, and impact.

Research and report Agents also need Trust Mode when sources conflict, citations are missing, files may be overwritten, attachments may be sent, or the output will be visible to customers. Confirmation cards let the owner decide whether the artifact is acceptable.

Boundaries and Poor Fits

Trust Mode does not replace business authorization, legal judgment, or security audit. It is a product boundary inside Agent workflows. It can route identifiable risk back to a human, but it should not be marketed as an absolute safety guarantee.

If a team will not define confirmation rules or review run evidence, Trust Mode becomes noise. If every action requires confirmation, automation disappears. The useful boundary is around irreversible, externally visible, or high-impact actions.

FAQ

FAQ: Does Trust Mode slow down Agents? It adds confirmation to critical actions, but it reduces the cost of wrong sends, wrong submissions, and mistaken permission changes. Low-risk steps can still run automatically.

FAQ: What belongs in a confirmation card? At minimum: action, source evidence, risk impact, suggested options, owner, expiry, and the accepted artifact or next handling step after confirmation.