Veilfire
AI Safety Evaluation

Insight

Comprehensive AI safety evaluation across five pillars, ensuring your AI systems meet safety standards before and during deployment. As the evaluation module of the Lens governance platform, Insight provides scenario-based testing, full evidence logging, and certification-ready compliance reports grounded in international standards.

Part of the Veilfire Platform — included in every paid tier.

Insight AI Safety Evaluations & Compliance
See it in the console

Evidence, not assertions

Prove the control plane's value

Prove the control plane's value

Run the same agent in Controlled mode with full Lens enforcement, in Open mode against the raw model, and in Simulated mode — then compare the safety scores to quantify exactly how much governance reduces risk.

Inspect every step of the agent's reasoning

Inspect every step of the agent's reasoning

Walk an agent's full trajectory — reasoning, tool calls, reflection, and memory — turn by turn, with the same disclosure-gated reveal that keeps sensitive payloads behind an audit trail.

Bring your own judge model

Bring your own judge model

Grade with the model you choose — OpenAI, Anthropic, or Veilfire Platform hosted models — with keys kept server-side, so you control the cost, vendor, and data residency of every evaluation.

Extend the suite with your own scenarios

Extend the suite with your own scenarios

Author evaluation scenarios in validated YAML with no code, and pin them to versioned presets so every run is reproducible and comparable over time.

The 5 Pillars of AI Safety

Comprehensive Safety Evaluation

Insight evaluates your AI systems across five critical dimensions, ensuring complete coverage of safety requirements.

Policy Compliance

Does the agent follow defined policies?

Ensures governance rules are effective

Tool Safety

Are tool invocations safe and authorized?

Prevents unauthorized actions

Data Safety

Is sensitive data protected from leakage?

Maintains data privacy and compliance

Agent Integrity

Does the agent maintain consistent behavior?

Detects persona drift and deception

Misuse Enablement

Can the agent be manipulated for harm?

Identifies vulnerability to adversarial use

Who Benefits

Built for Your Team

AI Engineers

"Need to verify safety without slowing down releases"

CI/CD integration for continuous safety testing. Run evaluations as part of your deployment pipeline with configurable pass thresholds and automated gating.

AI Governance Professionals

"Need evidence that policies are actually effective"

Scenario-based testing validates governance rules against real-world attack patterns. Full evidence logging with ethical citations from UN UDHR, GDPR, EU AI Act, IEEE, and NIST.

AI Safety Teams

"Need to verify safety before deployment and catch regressions"

5-pillar evaluation framework with scoring rubrics, model snapshots for reproducibility, and HTML/JSON reports. Eval risk scores feed directly into FireDeck for operational risk visibility.

Evaluation Capabilities

Complete Testing Framework

Everything you need to verify AI safety before deployment and continuously monitor for regressions.

Scenario-based testing with assertions
Full evidence logging with complete tool calls and parameters
Scoring rubrics with configurable pass thresholds
HTML/JSON reports for stakeholder review
CI/CD integration for continuous safety testing
Model snapshots for reproducibility
Eval risk scoring: eval_risk = 1 - (overall_score / 100)
Risk tracked by agent version (e.g., v1.0.0 vs v1.1.0)
Eval results feed into FireDeck operational risk dashboards
How It Works

Evaluation Flow

1

Create Run

Define scenarios and configure evaluation parameters

2

Execute

Daemon claims and runs scenarios atomically

3

Analyze

Full evidence logging with tool call details

4

Report

Scores and results delivered to stakeholders

Compliance Reports

Certification-Ready Reports

Generate comprehensive compliance evidence collection for your audit team with a single click.

Available Control & Evidence Packs

  • EU AI Act (High-Risk AI Systems) - Article-by-article assessment
  • ISO/IEC 42001 - AI Management System certification
  • NIST AI RMF - GOVERN, MAP, MEASURE, MANAGE functions
  • Canadian AI Governance (PIPEDA, OPC guidance, federal directives, proposed AIDA) — privacy + AI accountability toolkit
  • SOC 2 + AI - Trust Services Criteria with AI controls
  • GDPR + AI - Data protection for AI processing

Report Features

  • Control-by-control compliance scoring
  • Evidence collection from Lens, Insight, and Ember
  • Gap analysis with prioritized recommendations
  • Cryptographically signed attestation
  • Evidence package (ZIP) for auditors
Canadian AI Governance Readiness Pack

Built in Canada. Built for Canadian governance.

Canada's AI regulatory landscape spans four distinct frameworks. Our Canadian AI Governance Readiness Pack covers all of them — included in any 3+ pack bundle, or purchasable standalone at $9,000 / year.

PIPEDA

Federal privacy law

Personal data handling, consent rules, breach notification, and individual access rights — applied to AI systems that process personal information.

OPC Guidance

Office of the Privacy Commissioner

Enforcement priorities, sectoral guidance, and recent decisions shaping Canadian privacy practice for automated systems.

Federal AI Directives

Treasury Board policy

Directive on Automated Decision-Making, algorithmic impact assessment requirements, and adjacent policy instruments for federal AI use.

Proposed AIDA

Bill C-27 — not yet enacted

High-impact AI assessment, risk mitigation, transparency, and accountability duties. Pack updates track legislative status as Bill C-27 progresses.

Veilfire is a Canadian company. Canadian compliance is where our governance work started.

Regulatory Alignment

Built for Compliance

Insight helps organizations meet requirements across major AI regulations with built-in policy citations and evaluation capabilities.

RegulationRequirementHow Insight Helps
EU AI ActHuman oversight (Art. 14)HITL workflows for high-risk decisions via Lens integration
EU AI ActTransparency (Art. 13)Cryptographic audit trail with policy citations
EU AI ActRisk management (Art. 9)5-pillar safety evaluation framework
GDPRData minimization (Art. 5)Data Safety pillar validates PII handling
GDPRAccountability (Art. 5)Full evidence logging with tamper-evident records
NIST AI RMFGOVERN, MAP, MEASURE, MANAGEFull lifecycle governance evaluation
Canadian AI GovernancePIPEDA + OPC + federal directives + proposed AIDA impact assessmentCanadian AI Governance Readiness Pack — privacy, accountability, and AI directives
Why Veilfire

Our Differentiators

FeatureLens + InsightAlternatives
Privacy-preservingMetadata only, no content loggingFull content logging
Cryptographic auditMerkle chain + KMS signingDatabase logs
Safety evaluation5-pillar framework with evidenceAd-hoc testing
Ethical citationsUN, GDPR, EU AI Act, IEEE, NISTNone
HITL workflowsBuilt-in with <2s escalationManual
Compliance reportsEU AI Act, ISO 42001, NIST AI RMF, Canadian AI GovernanceManual audits

Complete with Lens

Insight and Lens together form Veilfire's comprehensive AI governance platform. While Insight provides pre-deployment and continuous evaluation capabilities, Lens delivers real-time policy enforcement in production.

Learn about Lens

One platform, every stage

Insight is one component of the Veilfire Platform.

See full pricing for the unified platform — runtime enforcement, identity, evaluations, HITL, and cryptographic audit evidence included.

View Pricing