DI-Guardrail · MosaicDM

Real-time integrity validation for AI-generated content.

AI systems produce responses that sound authoritative — but can contain unsupported claims, policy violations, or harmful content. DI-Guardrail evaluates every AI response deterministically, delivering a clear PASS, REVIEW, or FAIL verdict with consumer-ready explanations and a tamper-evident audit receipt.

Claim Validation Policy Enforcement Source Fidelity Harm Detection Contradiction Detection Audit Receipts

Evaluate Content POST /validate · Φ_intake

input_text * ?

The AI-generated text to check for accuracy, safety, and compliance

Vertical Preset ?

Sets domain, severity mode, and evidence mode for your industry

contract_type optional ?

Evaluation type — leave on auto to infer from content, or declare explicitly

domain optional ?

Industry context — affects how strictly certain checks are applied

severity_mode optional ?

How aggressively to enforce integrity rules

evidence_mode optional ?

Whether the AI response is expected to include source citations

source_ids ?

Reference document IDs the AI was supposed to answer from (optional)

source_contradicted ?

Signal that the AI may have contradicted the source material

request_id optional ?

Your tracking ID for this request

What will you get back?

Fill in the form and click Run Integrity Check. DI-Guardrail evaluates the content and returns a verdict, an integrity signal, consumer-ready explanations, and a full audit receipt — in milliseconds.

The verdict

PASS The content is clean — no integrity issues were detected. Safe to use as-is.

REVIEW Potential issues found — a human should review before this content reaches users.

FAIL A policy or safety violation was detected — this content should be blocked or corrected.

What gets checked

Unsupported or overconfident factual claims — statements the AI can't back up

Policy and compliance boundary violations — breaking your organization's rules

Harm and safety threshold breaches — content that could endanger or mislead

Source document fidelity — whether the AI accurately reflected its sources

Internal contradictions — statements in the response that can't both be true

Missing required evidence or citations when sources were expected

Scope violations — going further than what the source material supports

Every evaluation returns consumer-ready explanations you can surface directly to users, and a tamper-evident audit receipt — permanent proof this check happened and what the verdict was.