Shadow AI discovery
We surface AI in use across SaaS, code, cloud APIs, and network egress, then map each finding to an owner and a purpose. No survey required.
AI Governance
Know every model, Agent, and prompt. Then govern them.
Teams adopt AI faster than security can track it. Saint Fox finds the shadow AI, builds a living inventory, and turns policy into guardrails that run in your pipelines. Boards get a clear picture. Builders keep shipping.
AI inventoryIllustrative
AssetTypeRiskFramework
claude-3-opusFoundationMediumISO 42001
support-copilotAgentHighNIST AI RMF
contract-ragRAGMediumEU AI Act
analytics-llmInternalLowSOC 2
shadow-ext-llmUnsanctionedHighOWASP LLM
AI governance is the practice of knowing every model, Agent, and prompt your company uses, classifying the risk of each, and proving the controls work. Saint Fox runs that as a continuous operating program, not a one-time audit.
Ask a security team which AI their company runs, with what data, and who approved it, and most cannot answer. Surveys go stale the day they are sent. Saint Fox replaces the survey with signal: discovery reads from the systems where AI actually runs, so the inventory stays current and the risk picture stays real. Boards get a clear answer instead of a guess.
What we deliver
An AI governance program that holds up to questions
Six working parts, delivered as a managed service or as an advisory engagement that hands your team a running program.
Living inventory
A single record of models, Agents, prompts, data flows, and access. It updates from telemetry, so it reflects what is true today.
Risk classification
Each use case gets a risk tier aligned to the EU AI Act and your own thresholds. High-risk cases get a documented decision quickly, with the reasoning on record.
Policy as code
Approved patterns become guardrails in the pipelines developers already use. Most AI work proceeds without a manual review.
Continuous monitoring
We watch for new AI services, drifting permissions, and prompts that touch sensitive data, and we route the signal to the right owner.
Board and audit evidence
Two reports from one program: a Board summary in plain language, and an evidence pack mapped to ISO 42001, NIST AI RMF, and your controls.
The problem
Shadow AI is the new shadow IT, moving faster
A decade ago, security teams chased unsanctioned SaaS. The pattern repeats with AI, except the curve is steeper and the data exposure runs deeper. None of it shows up in a vulnerability scan or trips a firewall rule. The first time most teams hear about an AI system is when a regulator, an auditor, or an incident asks a question they cannot answer.
Foundation models called from code with embedded API keys
Browser extensions and SaaS features sending data to third-party LLMs
Autonomous Agents with standing access to production systems
Retrieval pipelines indexing sensitive documents
Fine-tuned models trained on data that should never have left the building
Prompt templates that leak customer records into logs
How it fits together
The AI governance stack
Signal flows up from the systems AI runs through, into a living inventory, then into controls and evidence. Each layer feeds the one above it.
Layer 05EvidenceBoard summaries and auditor packs from one dataset.
Board reportingAudit evidenceControl mappingsDecision log
Layer 04ControlsGuardrails as code, gates on high-risk actions.
Policy as codePrompt guardrailsHuman-In-Loop gatesOutput filtering
Layer 03Risk & policyEach use case tiered against frameworks and policy.
Risk classificationEU AI Act tiersData sensitivityApproval workflow
Layer 02InventoryA living record of every model, Agent, and prompt.
ModelsAgentsPromptsDatasetsOwners
Layer 01DiscoverySignal pulled from the systems AI runs through.
SaaS telemetrySSO logsCode scansCloud APIsNetwork egress
Framework alignment
Built on the standards your auditors already use
We reuse your existing ISO 27001 and SOC 2 controls where they apply, so the AI program builds on what you already run.
ISO 42001
AI management system. The backbone of the governance program.
ISO & AI audits ›EU AI Act
Risk-tier classification for every use case in scope.
Readiness program ›NIST AI RMF
Risk management functions mapped to your controls.
OWASP LLM
Top 10 risks for LLM applications, tested and tracked.
AI red teaming ›How we engage
Four steps from unknown to governed
01
Discover
Map AI in use across SaaS, code, APIs, and endpoints. Assign owners.
02
Classify
Risk-tier each use case against EU AI Act, ISO 42001, and your own policy.
03
Control
Deploy guardrails as code. Pre-clear approved patterns. Gate high-risk decisions.
04
Prove
Generate Board summaries and auditor evidence packs continuously.
Outcomes
What teams see in the first 90 days
100%
Shadow AI inventoried
Every model, Agent, and prompt mapped to an owner.
<2wk
First inventory
Usable posture readout within the first engagement sprint.
5x
Faster approvals
Pre-cleared patterns mean most AI work skips the committee.
1
Evidence source
Board and auditor packs from the same living dataset.
Planning targets, not claimed results. Actual timelines depend on environment size and access.
Related solutions
FAQ
Top 10 questions, frequently asked
AI governance covers the full lifecycle of every model, Agent, and prompt your organization uses: discovery, inventory, risk classification, policy enforcement, monitoring, and audit evidence. Saint Fox builds a single record of what AI exists, who owns it, what data it touches, and which controls apply.
We combine SaaS telemetry, identity and SSO logs, network egress data, code repository scans, and cloud API usage to surface AI services in use across the company. Each finding is mapped to an owner and a business purpose so the inventory reflects reality rather than a stale survey.
The program maps to ISO 42001 for AI management systems, the EU AI Act risk tiers, and the NIST AI Risk Management Framework. Where you already hold ISO 27001 or SOC 2, we reuse existing controls and evidence so the AI program extends what you have rather than duplicating it.
A first usable inventory typically lands within the first few weeks of an engagement, covering the highest-traffic AI services and Agents. Full coverage across code, cloud, and SaaS follows as integrations are connected. Timelines depend on environment size and access.
No. Policy is delivered as code and guardrails sit in the pipelines developers already use. Approved patterns are pre-cleared, so most AI work proceeds without a manual review. High-risk use cases get a fast, documented decision instead of a long committee.
Pricing is scoped to the size of your AI estate, the number of integrations, and whether you want advisory only or fully managed governance operations. Most engagements begin with a fixed-scope readiness assessment, then move to a managed retainer. Contact us for a tailored proposal.
You get a living inventory, risk classifications per use case, policy decisions with timestamps and owners, and control mappings to your target frameworks. Reports are written for two audiences: a Board summary and an auditor-grade evidence pack.
Yes. Agents are treated as identities with permissions, data access, and actions. We inventory their tools and scopes, apply least privilege, and add Human-In-Loop gates for high-impact actions such as transactions, deletions, or production changes.
We integrate with what you run. The AI inventory and control evidence feed your existing GRC, ticketing, and SIEM platforms through APIs. If you have no GRC tool yet, we can run the program and hand over clean, portable records.
Discovery and inventory run in Identify, guardrails in Guard, monitoring and evidence in Implement, and the quarterly Board readout in Enhance. The program is the VIGILE loop applied to your AI estate.
Ready to see your AI posture?
Book a readiness call with a Principal Engineer. We look at your AI estate and tell you what we find.