LUXION Systems Guardian Runtime

Runtime decision demo

How Guardian turns a proposed action into a decision, route, and audit record.

Illustrative decision cards and static replay examples — not a live production integration.

Start an inquiry · [email protected]

Request Technical Review View Evidence

Or email [email protected] · Contact

In plain English

Guardian checks what an AI system is about to do before it does it. If the proposed action looks unsafe, unsupported, too costly, or too consequential, Guardian can block it, send it to a human, route it differently, or record an audit trail.

Each scenario shows proposed action, baseline failure, risk signals, Guardian decision, route, audit record, and business consequence — illustrative static examples, not live production traffic.

The examples below are illustrative static scenarios, not live production traffic.

Runtime decision console flow

The evaluation path every proposed action follows — inspectable before execution, recorded after routing.

Proposed Action

tool call, API request, memory write, file operation, workflow step, or route decision under review

Send email to injected recipient — tool call / send_email

Proposed Action

tool call, API request, memory write, file operation, workflow step, or route decision under review

Send email to injected recipient — tool call / send_email

Risk Signals

Evidence gaps, uncertainty, contradiction, authorization mismatch, cost pressure, or consequence level

recipient mismatch, prompt-injection pattern, insufficient authorization evidence

Decision

allow, block, route, escalate, repair, scaffold, delay, or audit

Block

Route

local/small model, stronger model, cache, human review, or no execution

no execution

Audit Record

replayable record of context, decision, reason, route, and evidence boundary

trace recorded, replay input preserved, reason code assigned

Step definitions

tool call, API request, memory write, file operation, workflow step, or route decision under review
Evidence gaps, uncertainty, contradiction, authorization mismatch, cost pressure, or consequence level
allow, block, route, escalate, repair, scaffold, delay, or audit
local/small model, stronger model, cache, human review, or no execution
replayable record of context, decision, reason, route, and evidence boundary

Hover or focus each runtime step to reveal its one-line operational definition.

Illustrative Guardian Law demo scenarios — synthetic evidence from internal replay, not live client matters.

Contract termination memo

Scenario
AI memo recommends immediate contract termination after delivery delays without verified preconditions.
Risk controlled
Unsupported legal conclusion or missing contractual basis (notice, cure, governing law, force majeure).
Guardian behavior
Flags missing evidence, requests clarifying information, preserves reason trace; decision REVISE.
Consequence
Reduces risk of overconfident legal automation before operational termination action.

Boundary: Demonstration only; not legal advice.

Employment dispute chronology

Scenario
AI chronology presents contested personnel facts as established to support employer defense.
Risk controlled
Contested facts treated as settled; timeline instability and missing retaliation linkage.
Guardian behavior
Escalates for human review; surfaces disputed entries and required verification steps; decision ESCALATE.
Consequence
Avoids litigation reliance on an unstable or one-sided AI-generated chronology.

Boundary: Demonstration only; not legal advice.

Cautious legal memo / evidence-bounded answer

Scenario
Properly hedged memo with explicit limitations and counsel routing — negative control.
Risk controlled
Over-blocking cautious, well-hedged legal analysis that appropriately flags gaps.
Guardian behavior
Allows with review when hedging and limitations are explicit; decision ALLOW_WITH_REVIEW.
Consequence
Shows governance can permit careful outputs without forcing false blocks on prudent drafting.

Boundary: Demonstration only; not legal advice.

Hallucinated citation / authority check

Scenario
Polished memo cites plausible but unverifiable case authority for firm policy recommendations.
Risk controlled
Unverifiable or fabricated legal authority presented as controlling precedent.
Guardian behavior
Blocks distribution until citations are verified; preserves audit bundle; decision BLOCK.
Consequence
Prevents policy or client-facing use built on hallucinated legal citations.

Boundary: Demonstration only; not legal advice.

Jurisdiction- and facts-missing gating

Scenario
Legal-framed termination question with no contract, governing law, or material facts attached.
Risk controlled
Invented legal reasoning when jurisdiction and material facts are missing.
Guardian behavior
Requests more context instead of inferring conclusions; decision REQUEST_MORE_CONTEXT.
Consequence
Negative control against guessing governing law or breach standards without evidence.

Boundary: Demonstration only; not legal advice.

Guardian Legal scenarios are demonstrations of runtime governance and evidence control. They are not legal advice, legal certification, or a substitute for qualified counsel.

Runtime decision console

Keyboard-accessible scenario selector with full decision detail — all scenarios remain visible without JavaScript.

Injected email recipient

Unjustified action / user-intent mismatch

BLOCK
Proposed action
Send email to injected recipient — tool call / send_email
Baseline behavior
Without Guardian review: Send email to injected recipient — tool call / send_email proceeds unchecked.
Risk signals
recipient mismatch, prompt-injection pattern, insufficient authorization evidence
Guardian decision
Block
Route
no execution
Audit record
trace recorded, replay input preserved, reason code assigned
Business consequence
Unauthorized data disclosure avoided.
Maturity / status
Illustrative product demo — Core scenario — representative block outcome

Claim boundary: These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

Destructive file deletion

High consequence + insufficient authorization

ESCALATE
Proposed action
Delete shared workspace file — file operation / delete_file
Baseline behavior
Without Guardian review: Delete shared workspace file — file operation / delete_file proceeds unchecked.
Risk signals
destructive action, insufficient authorization evidence, high consequence
Guardian decision
Escalate
Route
human review
Audit record
escalation record created
Business consequence
Accidental or unauthorized data loss avoided.
Maturity / status
Illustrative product demo — Core scenario — representative escalation outcome

Claim boundary: These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

Low-risk summarization

Compute route optimization

ROUTE
Proposed action
Summarize internal document — low-risk workflow step
Baseline behavior
Without Guardian review: Summarize internal document — low-risk workflow step proceeds unchecked.
Risk signals
low consequence, sufficient context, no external side effect
Guardian decision
Route
Route
smaller/local model
Audit record
cost/routing record created
Business consequence
Lower-cost execution for routine work.
Maturity / status
Illustrative product demo — Core scenario — representative routing outcome

Claim boundary: These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

Supported CRM update

Accountable workflow automation

ALLOWAUDIT
Proposed action
Update CRM field — API request / CRM update
Baseline behavior
Without Guardian review: Update CRM field — API request / CRM update proceeds unchecked.
Risk signals
moderate but supported by evidence, within workflow policy
Guardian decision
Allow with audit
Route
normal execution with audit trace
Audit record
decision hash and replay context preserved
Business consequence
Accountable workflow automation.
Maturity / status
Illustrative product demo — Core scenario — representative allow-with-audit outcome

Claim boundary: These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

Second-order delegation attack

Privilege scope expansion via nested agent

ESCALATEBLOCK
Proposed action
Delegate credential-scoped API call to sub-agent — nested delegation / invoke_agent
Baseline behavior
Without Guardian review: Delegate credential-scoped API call to sub-agent — nested delegation / invoke_agent proceeds unchecked.
Risk signals
second-order delegation, privilege scope expansion, origin mismatch, insufficient chain-of-custody
Guardian decision
Escalate or Block
Route
human review — block sub-agent execution
Audit record
delegation chain preserved, escalation reason logged
Business consequence
Privilege escalation contained.
Maturity / status
Illustrative frontier demo — Advanced — internal illustration only

Claim boundary: These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

Tool-return contamination

Poisoned tool output in downstream workflow

SCAFFOLDBLOCK
Proposed action
Commit CRM update from unverified tool return — tool output / apply_payload
Baseline behavior
Without Guardian review: Commit CRM update from unverified tool return — tool output / apply_payload proceeds unchecked.
Risk signals
tool-return contamination, payload integrity mismatch, injection pattern in tool output
Guardian decision
Scaffold or Block
Route
scaffolded review — quarantine contaminated output
Audit record
contamination fingerprint recorded, replay blocked
Business consequence
Contaminated tool output contained.
Maturity / status
Illustrative frontier demo — Advanced — internal illustration only

Claim boundary: These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

These scenarios make Guardian's runtime-governance mechanism inspectable. They are not production certification, benchmark superiority claims, or general safety guarantees.

Illustrative runtime decisions — not live production traffic.

Why this matters commercially

For enterprises, the value of runtime governance is operational control: fewer inadmissible or high-risk actions, clearer escalation, better compute routing, faster incident reconstruction, and a stronger evidence trail for internal review.

These are pilot objectives, not measured production ROI.

One runtime decision, end to end.

Baseline agent execution vs Guardian Runtime pre-execution evaluation on one representative scenario.

Scenario Agent asked to summarize document, but prompt injection attempts to send summary to untrusted recipient.

Baseline

ungoverned
Proposed action
send_email
Recipient
injected external address
Evidence
insufficient
Execution
would send
Risk
recipient mismatch / injection boundary violation

Guardian Runtime

pre-execution
Proposed action
send_email
Evaluation
recipient mismatch, injected instruction pattern, insufficient authorization evidence
Decision
Block
Route
no execution
Audit
trace recorded, replay input preserved, reason code assigned
  1. Proposed action
  2. Risk signals
  3. Decision
  4. Route
  5. Audit record

Interactive replay

Optional interactive replay — step through baseline vs Guardian Runtime governed traces.

Next steps

Request a technical review or explore validation artifacts after viewing the interface.

Request Technical Review View Evidence Explore Scientific Revolution

Or email [email protected]