The Enforcement Layer of AI Governance
FERZ determines whether an AI action is permitted before it happens, blocks execution when governance conditions are not met, and preserves evidence for review.
Foundational research corpus, DOI-registered·Vendor-neutral published standard with a public conformance suite·Patent filings across the engine portfolio
Most AI governance today is monitoring. Monitoring can tell you what a system did. It cannot decide whether the action was allowed before it happened.
That decision requires evaluation before the action executes, and evidence of that decision that a third party can examine. This is authorization, and it is the layer most governance stacks are missing.
FERZ builds that layer. Every governed action is evaluated against codified rules before it takes effect, and every verdict is recorded as tamper-evident, replayable evidence. Everything on this page follows from that one commitment.
AI Governance Is Three Problems, Not One
Most governance tools tell you what an AI system did, or shape what it tends to do. FERZ answers a different question: is this specific action permitted, before it executes.
Knowing what a system did
Observability platforms, logging, and monitoring answer this problem. They describe behavior after it occurs.
Shaping what a system tends to do
Training methods, fine-tuning, and guardrails address this problem. They influence tendencies, not individual actions.
Deciding whether a specific action is permitted
This is the problem FERZ is built for: evaluating the action against codified rules before it executes, and recording the basis.
A complete governance stack needs all three. FERZ is architected to sit alongside the monitoring and alignment tooling an organization already runs, not to replace it.
How FERZ Works
Every governed action follows the same path, from proposal to examinable evidence. No engineering background is required to follow it.
Proposed Action
An AI system proposes an action, decision, or output.
Evaluation at the Boundary
The proposed action is evaluated against codified rules at the runtime authorization boundary, before execution or release.
Verdict
The action receives one of three verdicts: ALLOW, DENY, or ABSTAIN. ABSTAIN blocks execution unless an authorized human override is recorded.
Authorization Artifact
FERZ records the verdict and its basis as a tamper-evident, replayable authorization artifact for examination and audit.
Probably Compliant vs. Provably Compliant
When the auditor arrives, which would you rather be able to show? FERZ moves AI governance from confidence estimates and post-hoc explanations to a recorded authorization basis that can be examined after the fact.
Traditional AI Governance
- Confidence score
- Log entry
- Post-hoc explanation
- Limited replayability
FERZ Governance
- Recorded authorization basis
- Tamper-evident authorization artifact
- Replayable evidence
- Independent examination
Correctness is not permission.
The Regulatory Baseline Is Already Set
Organizations deploying AI in regulated settings are no longer choosing whether to govern. They are choosing whether their governance produces evidence.
Enforcement Active
Prohibited-practice enforcement is active, high-risk obligations are phasing in, and penalty exposure scales with global annual turnover.
PCCP Guidance Published
The FDA has published guidance on predetermined change control plans for AI-enabled medical devices. Change control is easier to defend when AI behavior is governed and examinable.
Examination Expectations
Financial-services AI deployments increasingly need defensible, reconstructable audit trails rather than after-the-fact explanations.
Sources: EU AI Act, FDA PCCP guidance. Specific obligations and timelines are addressed on the industry pages.
One Boundary, Multiple Engines
FERZ is built around a common runtime authorization boundary and a common tamper-evident record discipline. Different engines address different classes of governed AI behavior, including language, constraints, causality, behavioral adaptation, bias constraints, and distributed authority.
Before execution or release
Language and constraint engines evaluate proposed actions and governed outputs before execution or release.
Complex governance contexts
Causal, behavioral, and bias-constraint engines support more complex governance contexts.
Published architectures
Published architectures address distributed authority, rule governance, and examinable records.
For Technical and Diligence Review
FERZ argues that governance claims should be independently examinable. That standard applies to FERZ itself. Everything in this section is public, dated, and verifiable without contacting us.
The Five Tests Standard
A system that claims to authorize AI actions can be tested. The Five Tests Standard (5TS) is a vendor-neutral published standard defining five conformance tests. FERZ is architected around that standard.
Stop
Can the action be halted before any side effect?
Ownership
Who authorized it, and is that authority recorded?
Replay
Can the verdict be deterministically reconstructed and independently re-verified?
Escalation
Where does control transfer when policy cannot resolve the action?
Provenance
Is the origin of every input to the verdict established? Provenance is origin, not truth.
The current public conformance bundle verifies Stop, Ownership, Replay, and Escalation. Provenance is normative in 5TS v1.2.0; Provenance conformance is not yet assertable, and the specification says so.
Verify It Yourself
Run the tests
The 5TS repository includes public test vectors and a verifier. The conformance suite runs without access to any FERZ system.
Read the papers
The foundational corpus is published with DOIs on Zenodo. The core doctrinal claims on this site trace to dated, citable sources.
Examine the standard
The Five Tests Standard is vendor-neutral and independently adoptable. Its DOI record is the canonical citation.
What FERZ proves, and what it does not
An authorization verdict establishes that an action satisfied applicable constraints given the information available at decision time. It does not establish the factual correctness of the underlying AI outputs.
FERZ proves authorization, not correctness.
The Framework Behind the Engines
FERZ publishes the reasoning, not only the products. The governance section sets out the definitions, the positions, and the distinctions the architecture is built on.
Definitions
The core terms of deterministic governance, defined precisely: authorization, enforcement, replay, and fail-closed behavior.
Positions
The positions FERZ argues from, including that observability is not authorization, and that halt is not a verdict.
Distinctions
Eight adjacent categories, eight published distinctions: what FERZ is not, set out architecturally rather than asserted.
Where FERZ Fits
Three paths organizations consider, and what FERZ adds to each.
FERZ vs. AI Guardrails
FERZ vs. GRC Tools
FERZ vs. Internal Build
Intellectual Property as Infrastructure
FERZ pairs a deliberate intellectual property strategy with a published standard. The engines are the subject of patent filings, the governance architectures are published as defensive prior art, and the Five Tests Standard is released as a vendor-neutral specification.
Filed across the engine portfolio
Patent filings span the pathway engines, including international and United States applications. The public materials describe interfaces, not implementation mechanisms.
Published prior art
Published governance architectures establish defensive prior art and define higher-order governance structures without disclosing implementation detail.
The Five Tests Standard
The Five Tests Standard is published as a vendor-neutral specification so it can be adopted independently of any single implementation.
Governance for Your Vertical
FERZ is built for environments shaped by FDA PCCP expectations, EU AI Act obligations, and financial-services examination pressure. Specific deployments and compliance posture are scoped during engagement.
Governance for FDA Submission Contexts
Deterministic governance with tamper-evident, replayable records, built for environments shaped by FDA PCCP change-control expectations.
Distributed Authority and Examinable Records
Distributed authority over governing rules and decision records, with tamper-evident, replayable evidence. Deployment scope is engagement-dependent.
Replayable Decision Trails
Tamper-evident, replayable decision records and bias-constraint evidence for regulatory examination contexts.
Bias-Constraint Evidence for Underwriting
Bias-constraint evidence and tamper-evident decision records for state regulatory examination contexts.
From governance claims to verifiable evidence
FERZ is built around deterministic mechanisms, tamper-evident records, and published architectural standards designed to make governance an independently testable and verifiable property.
FERZ is raising. Investor materials are available at ferz.ai/investors.