MunichTech EXPO Grand Challenge Award Winner

Know what your AI can cause.
Prove what it actually did.

HAIEC is the evidence-bound assurance layer for consequential AI systems. Deterministic evaluation. Cryptographic evidence. Bounded decisions you can defend to a board, an auditor, or a regulator.

4–8 weeks after readiness · Non-production first · No telemetry required

The oversight gap

Your AI agents can move money, change systems, and trigger workflows. Your oversight was not built for that.

Autonomy is arriving faster than the ability to govern it. The blocker is not model capability. It is the inability to prove what the system can reach, under whose authority, and with what consequence.

82.9%

of organizations had AI projects delayed by security or compliance review

EMA / Protegrity 2026

84%

of enterprise AI pilots never reach production

Zapier 2026

88%

of AI agent pilots fail to ship to production

Industry research 2026

34%

of firms trust the actions their AI agents are taking

Forrester 2026

The consequence

When AI can take consequential action without a human in the loop, the executive team loses the ability to answer basic questions. What can it reach. Under whose authority. What is the maximum financial exposure. Where does the evidence stop. These are oversight questions, not security questions. And no existing tool answers them.

Executive oversight

Ten questions you can finally answer

These are the questions executives ask when AI starts taking action. HAIEC answers each one with source-backed evidence, and shows where the evidence stops.

01What can this AI agent actually reach, change, or trigger?

02Under whose authority is it acting, and where does that authority stop?

03What is the maximum financial exposure of a single run?

04Where does permission end and delegated discretion begin?

05What has actually been observed, versus what is merely code-capable?

06Which guardrails are configured, and which are actually enforced?

07What changed since the last review, and does it matter?

08Where is the evidence partial or missing, so we do not overclaim?

09Can we defend this decision to an auditor, a board, or a regulator?

10What is the smallest next step that would unlock a deployment decision?

The gap no one fills

Every tool you own sees a piece. Nobody sees the path.

SAST, IAM, GRC, observability, and AI platforms each solve a real problem. None of them connect intent, authority, capability, action, and consequence into one bounded decision. That connection is what HAIEC provides.

Existing stack

HAIEC

Scanner finds a capability in code.

Maps the capability to the identity, the credential, the observed action, and the consequence.

IAM shows what a role permits.

Shows what the role permits, what the code can reach, and what was actually done.

GRC records control status.

Produces evidence that the control was evaluated, decided, and enforced. Or reports it honestly as unknown.

Observability shows a span.

Connects the span to intent, authority, and a bounded consequence envelope.

AI platform tracks a model.

Tracks what the model can cause through the tools and credentials it holds.

Five planes, kept independent

HAIEC evaluates five authority planes and never collapses them: Requested, Policy Authorized, Effectively Granted, Code Capable, Observed. Gaps between planes are the findings. The reconciliation is deterministic and reproducible.

Requested

What the agent asked to do

Policy Authorized

What policy says is allowed

Effectively Granted

What credentials permit

Code Capable

What the code can reach

Observed

What actually happened

DAI

Delegated Action Integrity

Deterministic by design

No model judges another model. Same input, same output, same SHA-256 hash.

Executives do not need another black box telling them a different black box is safe. HAIEC's evaluation path contains no large language model. Every finding resolves to source evidence, and evidence that is missing is rendered honestly as unknown.

What deterministic means in practice

Re-run the same evaluation on the same evidence and you get the same result. Every conclusion is traceable to a source artifact. Every unresolved question stays visible as a frontier. Nothing is inferred silently. Nothing is scored with a hidden confidence number. The output is defensible because the method is defensible.

Database-level immutability

HAIEC's evidence store enforces write-once immutability at the database level. Even a compromised administrator account cannot alter past evidence. Every snapshot is SHA-256 hashed and parent-chained. Every bundle is Merkle-anchored with inclusion proofs.

What this unlocks

A board-ready record. An auditor-defensible decision. A reproducible benchmark that a regulator or an internal risk committee can verify without trusting the vendor.

Framework alignment

Already aligned with the rules you answer to

HAIEC maps findings to the frameworks, laws, and standards your compliance and audit teams already work in. One evaluation. Multiple framework outputs. No duplicated effort.

EU AI Act

Article 9 risk management · Article 13 transparency · Article 14 human oversight

SOC 2 Readiness

CC6.1 · CC7.2 · CC8.1 evidence mapping for change management and monitoring

NIST AI RMF

Govern · Map · Measure · Manage functions with subcategory traceability

ISO 42001

AI management system controls and evidence expectations

ISO 27001

A.12 operational procedures · A.14 system acquisition · A.16 incident management

Colorado AI Act

Impact assessment and risk management program evidence

NYC Local Law 144

Bias audit support for automated employment decision tools

GDPR

Data flow and automated decision-making evidence

HIPAA

Security rule evidence for AI systems handling protected health information

Security taxonomy

Mapped to the OWASP standards your security team already uses

HAIEC maps findings to both OWASP Agentic Top 10 and OWASP LLM Top 10. Your security team does not need a new vocabulary. The results land in the language they already work in.

OWASP Agentic Top 10 (ASI01 to ASI10)

ASI01Agent Goal Hijack

ASI02Tool Misuse and Exploitation

ASI03Identity and Privilege Abuse

ASI04Agentic Supply Chain

ASI05Unexpected Code Execution

ASI06Memory and Context Poisoning

ASI07Insecure Inter-Agent Communication

ASI08Cascading Failures

ASI09Human-Agent Trust Exploitation

ASI10Rogue Agents

OWASP LLM Top 10 (LLM01 to LLM10)

LLM01Prompt Injection

LLM02Sensitive Information Disclosure

LLM03Excessive Agency

LLM04Supply Chain

LLM05Data and Model Poisoning

LLM06Unbounded Consumption

LLM07System Prompt Leakage

LLM08Vector and Embedding Weaknesses

LLM09Misinformation

LLM10Output Handling

Executive value

Oversight, control, and financial exposure in one view

HAIEC is built for the people who own the outcome, not just the finding. The output is a decision artifact, not a findings dashboard.

OVERSIGHT

See every consequential action before it happens

A single view of what each AI system can reach, change, or trigger. Bounded by the exact scope you chose to evaluate. No hidden inference. No confidence score standing in for evidence.

CONTROL

Set the boundary, then prove it holds

Define the approved operating envelope for each agent. HAIEC shows whether observed actions stayed inside the bound, whether guardrails were evaluated and enforced, and where drift began.

EXPOSURE

Know the financial consequence before the decision

For each action family, HAIEC surfaces the maximum reachable consequence and the evidence that supports or contradicts it. Financial exposure becomes a bounded, reviewable figure instead of an open risk register line. Insurers and D&O underwriters increasingly ask for this evidence.

Evidence ingestion

Meets your evidence where it lives

Every ingest adapter normalizes input into the same standard evidence format. Malformed input is rejected explicitly with per-record failure detail, never silently dropped.

REST APIProgrammatic access for custom integrations

OTLP / OpenTelemetryTraces, metrics, and logs from runtime systems

Structured dataJSON, JSONL, CSV records

Source codeGitHub App scan or direct archive upload

IAM and policyRole bindings, policy documents, effective permission exports

Runtime observationOptional, explicitly authorized only

Telecom O-RANWG11 records, A1 policy, O1 data, RAN handover, slice flow, SBI transactions

Logs and alarmsStructured logs, alarm events, KPI measurements

Operating envelopeDeclared bounds, approval references, validity windows

Developer integration

Already installed where your developers work

HAIEC is available on GitHub Marketplace. Install in under two minutes. Findings post inline on the exact lines that matter. Compliance badges embed in READMEs.

GitHub Marketplace

HAIEC Compliance app. Three scan modes: Metadata Scan (free, no code access), Diff Analysis (scans changed files in every PR), and Full Repo Scan (deep scan with cross-file analysis). Inline PR comments. Verifiable compliance badges. Install at github.com/apps/haiec-compliance ↗.

Bounded engagement

A scoped engagement with four defined exits

No open-ended evaluation. No production access required. No telemetry required. The scope, the access, and the exit criteria are agreed before any work begins.

Scope

One AI-enabled system and one consequential action family. Bounded, reviewable, and agreed in advance.

Access

Minimum sufficient evidence. Non-production first. Runtime observation only when explicitly authorized and never required for the initial engagement.

Duration

Typically four to eight weeks after readiness. Fixed scope with a defined deliverable. No rolling extensions.

Data boundary

No production data, no user data, no sensitive business data required. Read-only access at every rung of the evidence ladder.

Exits

Proceed to pilot. Remediate and retest. Upgrade the evidence and re-evaluate. Stop or redesign. Four outcomes, all defensible.

Evidence access ladder

You choose how deep to go. Anything not connected stays unknown, and that stays visible in the report.

Source code, read-only

Static analysis of application code. No runtime access. Safe for non-production environments.

Identity and access configuration

Role bindings, policy documents, and effective permission exports. All read-only.

Policy and operating envelope

What was declared, what was authorized, and what bounds apply to each agent.

Provider IAM evidence

Cloud provider IAM exports where supported. Read-only, scoped to the evaluated system.

Runtime observation, optional

Only if explicitly authorized. Not required for the engagement. Non-production endpoint only.

Deliverables

What lands on your desk

Six artifacts. Every one is source-backed, bounded to the exact evaluated scope, and written for a decision rather than a finding queue.

AI Action and Access Map

Source-backed paths from agent and tool capability through implementation to consequential operations and effects.

Authority and Capability Reconciliation

Requested, Policy Authorized, Effectively Granted, Code Capable, and Observed. Kept independent, never collapsed.

Delegated Action Integrity (DAI)

Compares delegated authority against code capability, effective grant, and observed action. Shows whether the agent stayed inside what it was delegated to do.

Evidence Frontier

Where proof is established, partial, unavailable, or not assessed. Unknowns stay visible instead of turning into false confidence.

Compliance Twin

Versioned compliance state with drift detection. Reconstructs design, authorized, and observed states. Shows the first deterministic divergence.

Decision Package

Executive findings, technical proof, limitations, remediation priorities, and a bounded assurance record tied to the evaluated scope. Includes a verifiable Decision Receipt.

Credibility

Independently recognized. Already evaluated.

HAIEC has been tested on real autonomous systems and recognized by an independent competition jury. The evaluation method is already validated.

Recognition

MunichTech EXPO Grand Challenge Award

AI Action Path Assurance by Subodh Kc received the MunichTech EXPO Grand Challenge Award under Devpost rules at the Autumn Hackathon 2026, announced 26 September 2026.

Official listing· Newsroom· Scores· Announcement

Reference evaluation

Kestrel autonomous traffic protection

A live reference evaluation against a telecom autonomous agent. Frontiers kept explicit, contradictions preserved, authority planes reconciled. View the public report ↗

Code capable paths

1,322

Source-backed traces

100%

Frontiers explicit

You are looking at this before most of your peers are.

Evidence-bound assurance for consequential AI is a category that did not exist eighteen months ago. HAIEC won the MunichTech EXPO Grand Challenge Award in September 2026. The enterprises that adopt first define the reference benchmark. Everyone else inherits it.

The ask

One introduction. One system. One decision you can finally defend.

Open one consequential enterprise problem. We will scope a bounded engagement on one AI system, one action family, non-production, no telemetry, and deliver a decision-ready evidence package — typically four to eight weeks after readiness.

admin@haiec.com haiec.com/enterprise ↗