A Policy Decision Point for Every Action
POST an action envelope, get back a verdict. AegisAI’s PDP is deterministic and stateless on the fast path - the same input always yields the same decision and the same ledger hash.
Illustrative product view
What a policy decision point is
A policy decision point (PDP) is the component that answers one question: should this action be allowed? A policy enforcement point - say the Fintra MCP guard sitting in front of a tool call - asks the PDP per invocation, and the PDP returns an authoritative verdict. AegisAI implements this as a deterministic engine (action_pdp.py, decide_action) at POST /api/v1/aegis/pdp/decide.
The rule order
| Stage | Condition | Result |
|---|---|---|
| Tenant isolation | Action crosses tenant boundary | deny - risk 95 |
| Scope containment | Requested scope ⊄ granted scope | deny - risk 90 |
| Sensitivity: write | Action writes data | require_step_up |
| Sensitivity: sensitive | Touches sensitive data | allow_with_logging or require_step_up |
| Sensitivity: read | Read within scope | allow |
The verdict is one of seven values - allow, allow_with_logging, require_step_up, deny_recommended, human_review_required, contained, or blocked - returned alongside a risk score, Action Trust Score, trust band, enforcement view, and a reproducible ledger hash.
Deterministic and fail-soft
- Same input → same verdict → same ledger hash, so decisions are reproducible and auditable
- Stateless fast path: the decision does not depend on hidden session state
- Malformed input returns a conservative human_review_required - never a 500
- A failure degrades to caution, not to a silent allow
Actors it covers
- Humans and service accounts
- OAuth apps and CI bots
- AI agents and MCP tools
- All evaluated against the same tenant, scope, and sensitivity rules
Frequently asked questions
What is a policy decision point?
A policy decision point (PDP) is the engine that decides whether a requested action is authorized. An enforcement point calls it per action and acts on the verdict. AegisAI’s PDP evaluates tenant isolation, scope containment, and a sensitivity ladder, then returns a verdict, risk score, and reproducible hash.
Is the PDP deterministic?
Yes. The fast path is stateless and deterministic - the same action envelope always produces the same verdict and the same ledger hash. That reproducibility is what makes the decisions auditable and safe to replay when an auditor or engineer needs to verify why an action was allowed or denied.
What happens if the request is malformed?
The PDP is fail-soft: a malformed or ambiguous request returns a conservative human_review_required verdict rather than crashing or silently allowing the action. It degrades toward caution, so a bad input never becomes an unnoticed pass.
Does the PDP enforce the decision itself?
No. The PDP returns the authoritative verdict and records it; enforcement is gated by AEGIS_ENFORCEMENT_MODE, which defaults to simulate. Containment and revocation are simulated JSON, so the PDP is a decision fabric your systems consult rather than an actuator.
Stay in the loop
One practical finance briefing a week - new guides, checklists, and benchmarks.
Ask the PDP before you act
Send an action envelope and get back a deterministic, recorded verdict.
Talk to us