AI Governance & Responsible Use
Atlas Logic is an AI-native GRC platform. Our AI features run on Grok models (grok-4-mini base) hosted in Microsoft Azure AI Foundry, under Microsoft’s enterprise data-protection commitments. We govern all AI systems through documented policies, human-in-the-loop controls, and contractual restrictions on model training. Our program aligns with ISO 42001:2023 and the EU AI Act (August 2026 enforcement deadline).
Azure AI Foundry — grok-4-miniISO 42001 roadmapEU AI Act alignmentOWASP LLM Top 10Human-in-the-loopAI risk register
EU AI Act — August 2026 enforcement
Atlas Logic’s compliance-automation platform is classified as Limited Risk under the EU AI Act (not a high-risk system per Annex III). We maintain transparency obligations for AI-generated compliance narratives and audit summaries. ISO 42001:2023 certification is targeted for Q4 2027 as our primary AI governance attestation.
Human-in-the-Loop Controls
- All AI-generated compliance narratives are reviewed before being published as evidence.
- Agentic workflows require human approval for any action that modifies customer compliance data.
- AI risk assessments are flagged for human review above a configurable confidence threshold.
- No AI system makes final pass/fail determinations on certification controls without human sign-off.
Data Handling
- Customer evidence content is never used to train foundation models — contractually enforced under Microsoft’s enterprise data-protection commitments.
- AI processing occurs within Azure AI Foundry; prompts and completions are not used to train base models.
- AI inference is tenant-isolated — no cross-customer data leakage in prompts or context.
- All AI provider calls are logged with a SHA-256 hash-chained audit trail in Azure Log Analytics.
Security Controls (OWASP LLM Top 10)
- Prompt-injection protection: user inputs are sanitized before AI prompts are constructed.
- Output validation: AI responses are parsed and validated before rendering to users.
- Privilege minimization: AI agents run with least-privilege scopes — no write access without explicit user action.
- Insecure-plugin prevention: all AI tool integrations are reviewed and allowlisted before deployment.
Transparency & Explainability
- AI-generated content is labeled in the product — users always know when output is AI-assisted.
- Confidence scores and evidence citations accompany all AI compliance assessments.
- Customers can request a human review of any AI-generated compliance determination.
- ISO 42001 gap analysis is underway — certification targeted Q4 2027.
Active AI providers
AI provider routing & data agreements
Hosted inference — Azure AI FoundryMicrosoft Azure AI Foundry grok-4-mini (base model) | Active | AI audit narratives, control gap analysis, evidence classification, policy drafting, and risk generation. | Processed within Azure AI Foundry under Microsoft’s enterprise data-protection commitments · transfers covered by SCCs / DPF · customer data contractually prohibited from training foundation models. |