Govern AI before it acts.
Most AI safety today is post-hoc — outputs reviewed after the system has already acted. Noetik Governance moves the control point upstream: every request is evaluated against a deterministic policy stack before the model acts on it — returned as EXECUTE, HOLD or BLOCK.
What is PREEXEC?
A deterministic decision layer between the request and the model.
Three deterministic verdicts — EXECUTE, HOLD, BLOCK — returned before the model runs. EXECUTE lets a clear, compliant request through. HOLD returns a ClarityFeedback™ prompt asking the user to clarify or correct their own input, which is then re-scored. BLOCK stops a request that violates a non-negotiable obligation — legal, regulatory, safety — with a logged reason. An always-on Tier-1 safety floor (prompt-injection, abuse) fires regardless of the active policy.
Every decision — EXECUTE, HOLD, BLOCK — is written to a tamper-evident audit chain. Hash-linked, signed, non-repudiable. The chain is the documentation: compliance evidence is produced by operation, not assembled after the fact.
The reference engine for Pre-Execution Governance.
PREEXEC™ is the deterministic pre-execution engine. It runs inside the customer perimeter and evaluates each incoming request against the configured policy stack before the model acts — model-agnostic across closed-source LLMs, open-weight models, rule-based pipelines or hybrids. The same scoring can be applied to machine-generated text where a deployment needs it; the core job is governing the human input that sets the system in motion.
Self-hosted
Runs entirely inside the customer perimeter. No requests, prompts or audit entries leave the deployment boundary.
Low latency
Gating budget engineered for user-facing systems. The engine sits in the request path; it cannot be the bottleneck.
Model-agnostic
Evaluates requests bound for any upstream model. Switching the model does not require rewriting the policy stack.
Deterministic & replayable
Identical input plus identical policy version yields identical decision. Every run can be replayed end-to-end from the audit chain.
Tamper-evident audit chain
Every decision is hash-linked, signed and non-repudiable — suitable for regulatory inspection and forensic review.
Open evaluation harness
Corpora, runner scripts and result formats are documented and reproducible by the customer’s own audit function.
Built for the obligations that are coming.
AI deployments in regulated sectors no longer rely on goodwill audits. The EU AI Act enters its operational phase. ISO/IEC 42001 is the first ISO management-system standard for AI. UK regulators are issuing sector guidance with teeth.
High-risk system documentation
The audit chain produces the records required for Article 12 (logging), Article 14 (human oversight) and Article 17 (quality management) of the regulation.
AIMS controls
Policy stack and decision records map directly to the AI Management System controls of the first ISO standard for AI.
Govern · Map · Measure · Manage
Each RMF function has a corresponding artefact in the engine record: policy version, decision (EXECUTE/HOLD/BLOCK), decision rationale and audit chain entry.
Designed for environments where AI-assisted decisions are answerable.
- Insurance
- Banking
- Healthcare
- Public administration
- Pharma & life sciences
- Legal services