How BLEUN's permission model works
BLEUN grants each agent one of three graded levels of autonomy on a given task: read, where the agent may only gather information; propose, where the agent prepares a recommendation or draft for a human to review; and act, where the agent may execute directly within a defined boundary.
Irreversible or sensitive actions — payments, contractual commitments, external communications sent on your behalf — always require human approval regardless of the autonomy level otherwise granted to that agent. This model is described in more depth for organisations on the enterprise page.
Permission levels
Autonomy is set per agent and, for sensitive categories of action, per action.
| What the agent can do | When it is used | |
|---|---|---|
| Read | Gather and analyse information only, no changes made. | Research, monitoring and status checks. |
| Propose | Prepare a draft, recommendation or plan for human review. | Communications, offers, and any action with material consequences. |
| Act | Execute directly within a defined, bounded scope. | Low-risk, reversible or pre-approved routine actions. |
Data scoping, credentials and encryption
Access is scoped to what a given objective and task actually require.
Per-objective data scoping
Data gathered or produced for one objective is scoped to that objective rather than pooled across every objective in a workspace.
Credential and document handling
Credentials and uploaded documents are used only for the task that requires them and are not exposed to agents outside that scope.
Encryption
Data is encrypted in transit and at rest as a baseline control across the platform.
Audit logging
Agent actions are logged with their inputs, a reasoning summary and their outcome, so activity can be reviewed after the fact.
Deletion and export
Objective data can be exported or deleted on request; reach us via the contact page to initiate one.
Third-party AI model routing
BLEUN is provider-agnostic in how it routes requests to underlying AI models and limits what is sent to what the current task needs.
Incident reporting and responsible disclosure
If you believe you have found a security issue or want to report unexpected agent behaviour, contact us through the contact page. Provide enough detail to reproduce the issue and we will follow up on next steps and timeline.
What we do not do
- We do not sell personal data to third parties.
- We do not let an agent take a silent, irreversible action without an approval step.
- We do not grant agents unbounded autonomy outside the scope of the objective they were assigned.
- We do not claim a specific certification or completed audit that has not been verified.
Frequently asked questions
- Does an agent ever act without asking me first?
- Routine, low-risk work such as research and drafting can run within the autonomy you grant. Irreversible or sensitive actions — payments, submissions, external communications — require your explicit approval before they execute.
- What data does BLEUN send to AI models?
- Only the data an agent's current task requires. BLEUN is provider-agnostic in how it routes requests to underlying AI models, and scopes what is sent per task rather than exposing an entire objective's data at once.
- Can I delete or export my objective data?
- Yes, deletion and export of an objective's data is supported. Contact us through the contact page if you need help completing a request.
- Is BLEUN certified against a specific security standard?
- We do not claim any specific certification or completed third-party audit on this page. What is described here is our permission model and design intent; confirm current status directly with us before relying on it in procurement.