Pattern detected
IllustrativeProduction database migrations have been approved 8 times when requested by Release Bot during scheduled deployment windows.
Review recommendationKeep requiring approval
Adaptive engineIn development
The engine learns where judgment is needed.
BehalfID starts with the policies, permissions and authority requirements you define. Adaptive mode observes the decisions your team makes on top of them and turns repeated judgment into bounded recommendations.
Policy first. Learning second.
Decision historyIllustrative
1 similar decisions observed · approvals and declines both count as evidence
Release Bot asks to run a production database migration.
- Pattern confidence
- Insufficient evidence
- Handling today
- Approval required
- Policy
- Authoritative
Suggested handling
Not enough comparable decisions yet. Requests keep following the policy you defined, and the engine keeps watching.
Three states, in order, with a person at the gate.
Step 1
Observe
Decisions and their context are recorded. Nothing about runtime behaviour changes.
On by default
Step 2
Recommend
Repeatable patterns are surfaced as a proposed rule, with the decisions behind it.
Review in the console
Step 3
Enforce
An administrator enables a recommendation before it can affect a single decision.
Explicit opt-in
BehalfID does not silently expand an agent’s permissions. Learned patterns become bounded recommendations or administrator-enabled rules, and every one of them can be reviewed, audited or turned off.
What it learns
Patterns across allow, deny and approval.
Learning is not limited to declining things. The engine looks at the whole decision surface, and stays bounded by explicit policy in every case.
- Whether an action should be automatically allowed under an administrator-approved adaptive rule
- Whether an action should continue requiring approval
- Whether a recurring action is usually declined
- Which human role should review a request
- Which context is most relevant to a reviewer
- Whether an action differs materially from previously approved behaviour
- Whether a policy recommendation should be surfaced at all
Examples
What a recommendation looks like.
Reference content from a sample workspace. Each recommendation carries the evidence behind it and the administrator action it needs.
Repeated decline
IllustrativeRequests to expose public database ports have been declined 5 times.
Add explicit deny ruleDismiss
Reviewer routing
IllustrativeInfrastructure changes are consistently routed to the Security Lead.
Set default reviewerDismiss
Behaviour change
IllustrativeThis request differs from previously approved deploys because it adds a new external destination.
Continue requiring approval
Governance
Learning does not bypass administration.
- Confidence and evidence
- A recommendation always shows how many comparable decisions it rests on, who made them and how consistent they were. Thin evidence stays an observation.
- Administrator control
- Nothing learned changes runtime behaviour until an administrator reviews and enables it. Recommendations can be dismissed, narrowed or kept as approval-only.
- Audit history
- Adaptive rules are recorded like any other policy change: who enabled it, on what evidence, and every decision it has handled since.
- Rollback and disable
- An adaptive rule can be disabled instantly. Requests it used to handle fall back to the explicit policy underneath it.
Adaptive mode is in active development. Production includes adaptive-delegation surfaces for recommendations; the visuals on this page describe product direction and must not be read as measured accuracy or fully autonomous enforcement.
How we treat authority and audit