How HAIEC Works | Deterministic AI Action Assurance

From AI action paths to decisions you can defend.

Connect an authorized AI system or source. HAIEC identifies consequential capabilities and reachable paths, compares relevant policy, access, delegated authority and observed activity, then provides evidence-backed findings and scoped assurance results your team can inspect and act on.

Start with a finding. Understand the authority. Decide with evidence.

Scope an Enterprise POC Try HAIEC with Your AI Agent See a Real Assurance Report

Four-word goal for your connected agent

Run HAIEC AI Assurance

Use a compatible MCP client with scoped credentials and an authorized source. The AI operates the workflow; HAIEC determines the canonical result.

ONE CONSEQUENTIAL ACTION, THREE QUESTIONSExplore the workflow

SOURCE CAPABILITY

A support agent can reach a financial action.

A supported source path links an agent to a tool, a handler and a payment service. That is useful exposure evidence before any executed refund is established.

Supported finding

Potential recipient-changing action path in inspected code.

Business question

Which operating policy and recipient-choice authority apply?

Technical capability can be established without claiming the transaction occurred.

Four ways to start:MCP / IDE GitHub App Assurance API Workspace

Start where your team works

Choose your entry point. HAIEC owns the assurance result.

Use compatible MCP-enabled agents, selected GitHub sources, supported APIs or the HAIEC workspace. Every route ultimately depends on actual eligible sources and HAIEC’s canonical evaluator.

AI IDEs and agents

MCP connection

Best for teams whose agents already operate in a compatible client.

Connect a compatible client with a scoped key, check source readiness and request an eligible assessment.

Connect HAIEC MCP

Repositories

GitHub App

Best for source-level review of selected repositories.

Authorize selected repositories to examine supported code and consequential source-level paths.

Connect GitHub

Products and CI

Assurance API

Best for pipelines that need persisted evaluation artifacts.

Run permitted evaluations, poll supported jobs and retrieve exact recorded artifacts.

Explore the API

Organization teams

HAIEC workspace

Best for teams reviewing systems, findings and decisions together.

Select systems, connect available evidence and inspect related findings and decisions.

Open Workspace

The seven-phase customer lifecycle

A practical journey from question to accountable action.

These seven phases explain the customer experience. They are not seven mandatory gates before HAIEC can surface a consequential source finding.

04 / SEE

What consequential action paths can this agent reach?

HAIEC relates supported agent, tool, handler and downstream resources to potential consequences, including grounded indirect routes.

Explore System Constellation

What you gain

An inspectable action path and possible effect with source scope.

The evidence question

Code can demonstrate a reachable path without establishing completed execution.

Consequence analysis in practice

The path matters more than the individual components.

Explore an AI refund decision or a cross-customer information boundary. Both show how permitted tools can combine into a consequential operation worth investigating.

FINANCIAL ACTION

A support agent reaches a refund-recipient operation.

A connected code path may support a recipient-changing capability. Technical access and business authority to choose the recipient remain separate.

What HAIEC can find

Source-linked tool and handler reachability to payment logic.

Authority question

Is the recipient fixed by policy or selected under a genuine delegated choice?

Next accountable step

Inspect the relevant rule and authorized choice before claiming an observed transaction.

Explore Action Mapping See Kestrel Constellation

Five independent evidence planes

Five different truths. A more defensible decision.

No single source is permitted to prove every claim. A supported code finding, applicable policy and relevant runtime event can each contribute independently.

REQUESTED

What did the user or workflow request?

A request can define the target task, intended action and relevant scope.

What it establishes

The authenticated request supplies evidence of task intent.

What stays distinct

Intent is not policy authorization, effective access or delegated business choice.

Delegated Action Integrity (DAI). A credential can allow an operation without establishing who was entrusted to select the consequential outcome. DAI is a cross-cutting authority comparison, not a sixth evidence plane. Explore DAI

Real value at each evidence level

Find, evaluate, verify and follow through.

Identify

Relevant exposure

Supported potential agent, tool and handler path with a consequential effect worth investigating.

Compare

Deterministic assurance

Relevant policy, grants, delegation and qualified observations contribute to the scoped evaluation.

Inspect

Verifiable artifacts

Where issued, reports and receipts bind relevant evidence and exact evaluation identity.

Operate

Connected follow-through

Review decisions and monitored changes where integrated. Customer-side blocking needs its own effect evidence.

Agent-native onboarding

Your agent guides the workflow. HAIEC evaluates the evidence.

Configure HAIEC in a compatible MCP-enabled IDE or agent. A scoped private key and an authorized eligible source are required for actual evaluation.

Connect HAIEC MCP Inspect a Sample Report

HAIEC MCP WORKFLOWScoped connection

You: Run HAIEC AI Assurance

→ haiec_preflight
  Check eligible system and source access

→ Source analysis
  Identify consequential tool and handler paths

→ Eligible assurance evaluation
  Compare relevant authority and evidence

→ Inspect the report
  Findings, exact scope and next decisions

The AI operates the workflow.
HAIEC owns the bounded evaluation.

Endpoint: https://www.haiec.com/api/mcp. Configure authorization in the client.

Sources and technical foundations

Start with the evidence you have. Enrich the question as needed.

HAIEC can use selected source code, identity, policy, provider access, connected runtime and declared inputs for their relevant claims. Not all sources are mandatory for an initial source finding.

Source analysis

Repository actions, tool relations and reachable code.

Identity and access

Principal and access configuration.

Policy evidence

Operating policies and constraints.

Provider IAM

Effective provider grants where supported.

Runtime evidence

Events and effects observed by connected sources.

Declared evidence

Relevant owner-confirmed and uploaded records.

Explore the technical engines supporting the journey

Source Analysis

Static findings and capability paths.

Evidence Core

Source scope, lineage and related context.

Topology Projection

Supported System Constellation relationships.

Assurance Decision Engine

Deterministic evaluation for eligible claims.

Package and Receipt Layer

Persisted reports and integrity context.

Audit Orchestrator

Coordinates supported evidence producers.

Explore Platform What Is HAIEC?Explore Sample Map Deep-Dive Methodology How LLMs & Agents Work

Frequently asked questions

From source findings to decisions with evidence.

What does HAIEC analyze first?

Begin with an authorized AI system, an eligible source and a consequential question. Supported source analysis can find potential action paths before every provider is connected.

Can HAIEC find issues using repository evidence alone?

It can identify source-supported possible capabilities and derived paths in analyzed code. Confirming actual runtime execution is a separate evidence question.

Do I have to connect all five evidence planes?

No. They answer independent questions. The selected claim determines which facts are necessary for a stronger conclusion.

How does HAIEC differ from an AI security scanner?

It connects relevant component findings to consequential action paths, policy, effective access and delegated choice, then supports deterministic evidence-backed assurance.

Does tool permission authorize a business decision?

Not always. Business discretion can be different from technical access. A rule-grounded choice or authorized bounded delegation can establish specific decision rights.

Does another AI model issue the verdict?

No. A configured AI assistant can guide the workflow, while HAIEC’s deterministic evaluator owns the canonical bounded conclusion.

What do ALLOW, REVIEW and BLOCK mean?

These dispositions apply within their actual evaluated scope and contract. They do not by themselves assert universal safety, certification or target-side enforcement.

Does HAIEC continuously monitor or stop all agents?

Monitoring depends on supported connected sources, and prevention needs an installed, separately verified customer-side intervention path.

How can I verify a report?

Review the persisted source identity, evaluated policy, evidence and any available integrity receipt. Verification is bounded by that specific evaluated basis.

How can I begin an enterprise evaluation?

Scope an Enterprise POC, or connect through MCP, authorized GitHub, APIs or the workspace. Begin with one consequential system and a question your sources can answer.

High assurance in every consequence

Know what your AI can cause. Decide with evidence.

Understand the agent actions your sources can support, compare authority behind consequential choices and move AI forward with a defensible next decision.

Scope an Enterprise POC Try HAIEC via MCP

A focused start

One AI system. One consequential question.

Find the path, understand the relevant authority, inspect the evidence and determine the next accountable action.

See an Assurance Report