Beyond Zero
Machine-speed dynamic authorization for the AI era. Blends static ReBAC policies with AI-driven risk evaluation — securing both humans and AI agents without overburdening users.
Aligned with Google's Beyond Zero architecture (Valente & Zalewski, 2026). Available as an add-on for Professional and Enterprise plan subscribers.
The Five Pillars
Beyond Zero extends Zero Trust from the application boundary to individual actions on individual resources.
Resource/Action-Based Security
Every authorization decision is made at the individual resource and action level — not at the application or tool boundary. True per-request evaluation at machine speed.
Blended Static + Dynamic Security
SpiceDB relationship-based policies (the static floor) combined with AI-driven risk evaluation (the dynamic ceiling). Static policies remain auditable; dynamic controls add contextual intelligence.
Automatically Enriched Context
Pre-computed accessor profiles, resource sensitivity, work assignments, device posture, and behavioural baselines — all available at sub-millisecond access time.
Automated In-Depth Investigation
AI-powered investigations triggered by risk signals. Correlates access patterns, detects enumeration and exfiltration, and produces GDPR Article 22 compliant decision explanations.
Challenges and Containments
Granular, context-aware challenges (justification, security key, manager approval) and escalating containments — from rate limiting to full access denial.
Supported Features
All features are available when Beyond Zero is enabled on your plan.
Dynamic Authorization Engine
Fast Path Evaluator
In-process compiled policy evaluation in < 3ms p99. Zero synchronous network calls on the hot path.
Staleness Circuit Breaker
Detects stale risk profiles and elevates risk scores until event intake catches up.
Slow Path Investigations
AI-driven pattern analysis via DeepSeek. PII-scrubbed, prompt-isolated, GDPR-compliant decision DAGs.
Policy Dry-Run
Simulate new policies against historical data before enabling. See impact without risk.
Contextual Caveats (ABAC + ReBAC)
SpiceDB CEL caveats evaluated at check time — time-of-day, location, auth strength, risk score, work assignment.
AI Agent Security
Agent Identity Registry
First-class agent identities with declared capabilities, bounded sessions, and controlling human attribution.
Delegation Chain Validation
Multi-hop agent delegation with scope intersection. Permission can only narrow, never widen, across hops.
Intent Verification
Structured JSON intent declarations (not free-text) validated against resource access. Prevents prompt injection and scope creep.
Ambient Authority Prevention
Agents never inherit the full permission set of their controlling human. Permissions are the intersection of human + agent + task scope.
Human Intent Alignment
Vocabulary-based tokenisation of human task declarations. Agent actions verified against the human's stated objective.
Containment & Response
5-Level Containment Model
None, Rate Limited, Sensitivity Restricted, Read Only, Full Containment. Automatic escalation/de-escalation.
Policy-Linked Challenges
Justification, security key touch, manager approval, controlling human confirmation, re-authentication, rate reduction.
Break-Glass vs Level 4
Break-glass freely bypasses L1-L3, but L4 (suspected compromise) requires FIDO2 + peer counter-signature.
CAEP Immediate Containment
Session revocation events trigger Level 4 containment within 2 seconds. No investigation delay.
Autonomous Containment
High-confidence investigations (>90%) apply containment without human approval. GDPR Article 22 explainability guaranteed.
Device & Identity
Device Posture Attestation
MDM-sourced trust levels (managed, compliant, encrypted, hardware-bound). Jailbroken = always untrusted.
NHS Smartcard Binding
DPoP proof with x5t#S256 certificate thumbprint validated against NHS Spine CA. Hardware root of trust.
Federated Identity (RFC 8693)
Token exchange with Entra ID, CIS2, custom OIDC. Canonical subject format, namespace scoping, health monitoring.
Continuous Verification
Mid-session re-evaluation on every request. Device posture degradation, risk threshold crossing, CAEP revocation — all trigger immediate response.
JIT Elevation & Clinical Safety
Just-in-Time Privilege Elevation
Time-bound access via SpiceDB expiring relationships. Auto-approve for low-risk, auto-revoke on expiry. Zero manual cleanup.
Locum & Agency Access
Pre-configured elevation profiles: max duration, approver role, cooldown period, renewal limits.
Offline PDP (Clinical Continuity)
Local PDP sidecar serves cached decisions during network failure. Clinical safety outweighs absolute consistency.
Clock Drift Protection
Edge nodes validate clock against central API. Configurable drift threshold (30-120s) for rural NHS environments.
Observability & Compliance
28+ Prometheus Metrics
Evaluation latency, challenge resolution, containment active count, anomaly detection, cache hit rates, circuit breaker state.
GDPR Article 22 Explainability
Deterministic decision DAGs for every autonomous containment. Template-based human-readable explanations within 72 hours.
GDPR Erasure (Pseudonymisation)
SHA-256 salted hashing, 7-table update, Redis cleanup, irreversible salt deletion. Preserves audit trail structure.
DSPT Control Mapping
Every Beyond Zero capability maps to specific Data Security and Protection Toolkit controls.
Activity Window
Sliding window of last 100 actions per accessor. Pattern detection for enumeration, exfiltration, and privilege probing.
Pre-Built NHS Caveat Templates
SpiceDB CEL caveats evaluated natively at permission-check time. Attach to any relationship.
nhs_working_hoursAccess restricted to 07:00-19:00 UK time
nhs_uk_onlyAccess restricted to GB-located devices
nhs_smartcard_requiredRequires Smartcard or FIDO2 authentication
nhs_risk_boundedDenied when risk score exceeds threshold
nhs_assignment_scopedRestricted to current work assignment scope
nhs_device_trust_requiredRequires managed, compliant device
Beyond Zero is an Add-On
Beyond Zero is available as an add-on option for Professional and Enterprise plan subscribers. It operates alongside AuthHub's core ReBAC engine — the static SpiceDB floor continues to function independently. Beyond Zero adds the dynamic ceiling on top, and can be enabled or disabled at any time via a feature flag with zero impact on your existing authorization flow.
Performance Guarantees
Beyond Zero is architecturally decoupled from the SpiceDB critical path. It never degrades your baseline 100K RPS.