Listen to intro Stations Evidence Lenses Compliance Media TRY THE DEMO
Runtime guardrail control plane

Every AI decision
proven before it runs

A runtime guardrail that sits between your people and your AI models. Pick a request, watch ten stations inspect it, read the cryptographic verdict.

1Pick a request
→
2Watch the rail
→
3Read the verdict
evidence.chain
SCROLL
The Gauntlet
Ten stations.
Six before the model. Four after.

Every AI request passes through a pre-flight and post-flight inspection chain. Nothing reaches the model unchecked. Nothing leaves unverified.

S01PRE
Identity & Purpose
Caller authentication, intent classification, prompt-injection detection
S02PRE
Data Classification
Sensitivity tagging, token-level redaction before the model sees the payload
S03PRE
Personal Data & Secrets
Credential scanning, PII detection, secret masking
S04PRE
Manipulation
Adversarial prompts, jailbreak patterns, social engineering vectors
S05PRE
Model Authorization
Tier-based model access, deployment region enforcement
S06PRE
Spend & Rate
Budget gates, token-rate throttling, cost attribution
MODEL EXECUTES
S07
Grounding
Factual verification, data-sovereignty rerouting
S08
Regulated Conduct
Fair lending checks, bias detection, prohibited advice screening
S09
Disclosure
Human-in-the-loop escalation for consequential decisions
S10
Human Oversight
Final accountability gate, audit seal
LIVE VERDICT FEED
0
ALLOWED
0
BLOCKED
0
OTHER
Immutable Proof
SHA-256
hash-chained evidence

Every verdict, hold, release, policy deployment, halt, and resume signature — appended to one cryptographic chain. Tamper-evident by construction.

VERDICT — BLOCKED
S03: Prompt injection detected in forwarded email body
prev 0000000000000000…
hash a1f7c93d84e2b617…
VERDICT — REDACTED
S02: Client identifiers masked — 847 tokens sent to model
prev a1f7c93d84e2b617…
hash 3e8b1a6fc9074d23…
VERDICT — REROUTED
S07: Confidential treasury data rerouted to EU private deployment
prev 3e8b1a6fc9074d23…
hash 7d4e2f18a3c19b56…
HOLD
S09: Credit decline held for Accountable Reviewer
prev 7d4e2f18a3c19b56…
hash b29c5d0af8e61c47…
HALT
All AI traffic halted by Duty Officer — dual-sign resume required
prev b29c5d0af8e61c47…
hash e61f8a47d2b93c08…
CHAIN INTEGRITY
0VERIFIED
0
Records
0
Tamper alerts
v2.4.1
Policy ver.
Sealed
Attestation
EXPORT FORMAT
JSON
CSV
PDF
Seven Outcomes
Not just allow/block.
Nuance at machine speed.
Allowed
Redacted
Rerouted
Queued
Held for review
Blocked
Halted

Policy-pinned verdicts

Every verdict cites the exact policy version and SHA-256 hash it was decided under. No ambiguity about which rules applied.

Human-gated controls

Holds require a named human reviewer. Halts require dual-signature resume. The chain records who acted, when, and under which authority.

Executive Lenses
Four views. One ledger.
Complete accountability.

The same evidence chain read through the lens that matters to each executive. No separate reports. No reconciliation.

CFO
Financial
Spend under governance, modelled exposure avoided
  • Token spend vs. budget
  • Exposure avoided by controls
  • Cost of the control plane
  • Editable assumptions
CAIO
AI Governance
System register, risk tiering, regulatory crosswalk
  • AI system register
  • Risk tiering matrix
  • EU AI Act mapping
  • NIST AI RMF alignment
CTO
Architecture
Measured latency, fail posture, policy as code
  • Per-station latency (p50/p99)
  • Fail-closed posture
  • Policy version control
  • Deployment topology
CIO
Assurance
Chain verification, tamper test, sealed attestation
  • Control coverage
  • Chain integrity check
  • Evidence export
  • Separation of duties
Regulatory Crosswalk
Built for the frameworks
that matter.

Mapped by station, not bolted on after. Anchors for validation, not legal conclusions.

EU AI Act
High-risk AI system obligations
NIST AI RMF
AI Risk Management Framework
SR 11-7
Model Risk Management
BCBS 239
Risk Data Aggregation
ISO 42001
AI Management System

See it run.

Pick a scenario, watch the ten stations inspect it, read the verdict. Takes two minutes.

Try the demo