Executive Assurance Summary

Plain-English narrative for leadership. Evidence-backed conclusions only. No technical jargon, no LLM-generated prose.

Challenge Alignment

How HAIEC aligns with the enterprise telecom challenge criteria. We do not self-grade; we show what was evaluated and where the evidence sits.

Data Intake & Source Binder

What was received, what each source can establish, and what remains unproven. This is the Environment Binder — it determines the limits of every claim in this report.

Intake Adapter Layer

All evidence sources HAIEC can accept. Every adapter normalizes to a canonical envelope, then runs the detection engine. Unknown formats are quarantined, never silently dropped.

✅ Available now (14 adapters)

⏳ Coming soon

Telemetry Signals

OpenTelemetry traces, metrics, and logs gathered from the evaluated surfaces via bounded collection. Signals are correlated to findings; baseline comparison runs over the collected window. Gathered, not streamed

Estate Constellation

Interactive map of the AI estate. Node size = reach. Brightness = evidence confidence. Ring = trust domain. Click any node to inspect. Evidence discipline

Agent Cards — Enterprise Telecom Estate

Each agent answers 9 questions about identity, capability, authority, and gaps. Click any card to expand. Evidence discipline

Capability Ladder

Declared → Code-Capable → Effectively Reachable → Indirect → Observed. The gap between declared and observed is the derived capability surface. Why the gap matters

Five-Plane Authority Model

Requested → Policy Authorized → Effectively Granted → Code Capable → Observed. Each plane requires its own evidence. No plane is populated from inference. Plane discipline

Agent Capability Chain

For each observed action: what capability was required, what was declared, what was observed, and what is the gap. Gap types

Delegated Action Integrity (DAI)

For each consequential choice: did the system stay inside what it was delegated to do? Four facts compared across 9 dimensions. Four facts

Supply Chain Influence

Packages that can change AI behavior if their scope changes. Direct and transitive dependencies with scope-change risk. Scenario rows

Detection Catalog

Every rule in the deterministic detection catalog. Each rule is evidence-bound, has hard negatives, and maps to real published framework IDs. Deterministic core

Framework Mappings — Real IDs, Not Badges

Every finding maps to real published framework IDs. Framework mapping ≠ certification. Standards inform; HAIEC evaluates.

Remediation Playbooks

When HAIEC finds an issue, this is what to do next. Playbooks are recommendations — never auto-applied. You decide. Recommendation discipline

Static Security Findings

Pinned Semgrep rulepack covering prompt injection, RAG poisoning, data leakage, tool abuse, and AI governance. Deterministic pattern matching. Runtime verify ready

AL0 — Baseline Forensic Report

Functioning fault management scenario. No injected threat. All authority planes established. No divergence.

AL1 — Degraded Forensic Report

Reliability / decision-quality degradation. First divergence detected at Step 4 (Action). All subsequent steps inherit this gap.

AL2 — Compromised Forensic Report

Cross-slice isolation violation + audit-trail tampering. Tampering detected on State Observation source. Independent corroboration survives.

Divergence & Diagnosis

What changed, why it changed, and what it impacts. Deterministic cause tree — no root cause claim without causal evidence. Earliest ≠ root cause

Guardrail & Remediation

What guardrails were applied, did they mediate the action, and what changed. Guardrail configured ≠ guardrail enforced. Ladder states

Canary — Guardrail Drift Detection

Fixed deterministic probes run hourly. Refusal rate is compared across windows to detect guardrail behavior drift. Deterministic probes

Alerts & Live Warning

Real-time push when a CRITICAL finding fires. Banner on the dashboard, toast on HIGH. Email and Slack dispatch in parallel. SSE + polling fallback

CI Release Verification

Every release gated by deterministic capability diff, envelope check, and forensic gate. Tamper-evident evidence packages — HMAC-signed receipts over Merkle-rooted evidence bundles.

Material Stories

The findings that matter most, composed deterministically — no ranking model, no LLM prose. Every story carries its evidence refs, limitations, and what would close its frontier. All remaining findings stay available below.

Four-Axis Reading

Every item is scored on four independent axes — never collapsed into one score. The gate is primary; the axes underneath explain why. Deterministic

AI Agent Path Map — Given vs Derived

Every AI agent has two paths. The given path is what you told it to do — config, policy, roles. The derived path is what it can actually do — what the code can invoke, what credentials can reach, what shared services connect to. The difference is your hidden capability surface. We don't guess

Deep Code Path Analysis

The code path HAIEC read. Every edge is a real code location — no inference, no guessing. The frontier shows what was not analyzed and why. Qualified extraction

Semantic Relation Graph

HAIEC understands 100+ typed relations — core graph, telecom, delegation, forensic, supply chain. Every relation requires evidence. A name match alone never qualifies one.

Action Witness — 5-Phase Progression

Every action has five phases: REQUESTED → AUTHORIZED → ACCEPTED → APPLIED → CONFIRMED. Each phase needs its own evidence. HAIEC never assumes the next phase happened — it only shows what's proven. Phase authority rules

Evidence Frontier

An evidence frontier is where proof stops. It's not a failure — it's exactly what we can prove and what we'd need to prove more. Every frontier shows what would close it.

Trust Boundary Map

Where the system crosses its trust boundary. Every crossing shown is real observed or connector-bound traffic — not a guess. Internal→external crossings are highlighted.

Network Slice Map

Network slices are supposed to be isolated. If traffic crosses a slice boundary, it's a violation. HAIEC tracks every cross-slice flow with evidence.

Tower Forensics — Handover Chain

Which tower did the action reach? How many towers were touched? HAIEC tracks every cell handover with evidence from RAN handover intake.

Cross-Language Capability Bridge

The same dangerous operation has different names in different languages. HAIEC normalizes them to the same canonical capability — so the finding is the same regardless of implementation language.

Also normalized: CRYPTO_INIT · DATABASE_QUERY · FILE_SYSTEM_WRITE — one canonical finding per capability, per language.