Your board may ask. "What AI is running here, and what data is it seeing?"
Leadership teams may not be able to say quickly which AI tools are in use. An AI inventory can surface tools that leadership did not know were in use, some of them touching customer data. The Preside AI Governance™ assessment produces the inventory, the data-flow map, a framework gap analysis, and a remediation roadmap. For fixed-scope AI work, see the two-week AI Technical Readiness Initiative and the six-week AI Security and Compliance Initiative.
What shadow AI looks like
Four example scenarios
Invented examples of the kinds of tools an AI inventory can surface in a mid-market environment.
A rep pastes a customer's full P&L into a public AI chatbot to draft a proposal
Customer-confidential financials enter a third-party LLM. Some vendor terms allow training on data unless the user opts out. The customer may have to be notified under the terms of its contract.
Tools of this kind: public AI chatbots used on personal accounts
Production source code flowing through IDE-embedded copilots
Developers using AI code completion that sends code context to the vendor. Repo contains customer-specific configurations and embedded API keys. Vendors differ in how they disclose training-data usage.
Tools of this kind: AI code assistants built into developer tools
Support agents using AI summarizers on full customer ticket histories
An installed browser extension is summarizing customer email threads from your help-desk tool. PII flows through a vendor nobody on the IT side has vetted. The extension's terms permit "anonymized analytics".
Tools of this kind: browser-extension AI assistants and AI meeting-note tools
Several generative-AI subscriptions, overlapping use cases, no audit trail
Departments bought their own AI tools on expense reports. Some overlap. IT has no complete list. The next SOC 2 cycle may ask about all of them.
Tools of this kind: AI writing, design, voice, and video generators
Why it has become a board topic
Frameworks now exist and boards are asking about them.
AI tools have spread through many departments in many organizations. Often there is no single "AI deployment". There is a steady buildup of tools, each adopted because it solved someone's immediate problem.
Now the questions arrive from above and across. The board wants to know what's running. Counsel wants to know what data is exposed. The CISO or security lead needs a documented inventory and control posture before the next incident forces a less measured answer. An auditor may ask about AI controls in the next SOC 2 cycle. The starting point for all four is the same record: a complete inventory, a data-flow map, and a mapping of controls to the frameworks you choose to follow. Preside maps practices to frameworks. That is not legal advice, and it does not establish compliance with any law. Counsel should confirm what applies to you.
Released January 2023. A voluntary framework for managing AI risk.
In force since August 2024. Obligations apply in phases, and the schedule may change. Confirm current dates with counsel. It can reach companies outside the EU. Penalties for prohibited-AI violations can reach 7% of global annual turnover.
SOC 2 criteria such as Security, Confidentiality, and Privacy can cover AI use, and ISO 27001 controls can too. Ask your auditor how AI will be treated in your cycle.
Examples of a changing patchwork, summarized briefly. Illinois HB 3773 (in force Jan 2026) addresses discriminatory effects of AI in employment. NYC Local Law 144 requires a bias audit of automated hiring tools. Texas TRAIGA (in force Jan 2026) restricts certain uses of AI, including behavioral manipulation and certain deepfakes. Colorado SB 26-189 (effective Jan 2027) sets notice, adverse-decision disclosure, and human-review duties for automated consequential decisions. California SB 942 (operative Aug 2026) sets watermarking and detection-tool duties for large generative AI providers. Utah's AI Policy Act requires disclosure of generative AI use when a consumer asks, and in some higher-risk interactions. Scope and dates change. Confirm the current text and its application to you with counsel.
The assessment
Four dimensions, one record
Designed to produce one record that auditors, counsel, and boards can each work from: a single complete inventory of AI in your environment with risk and remediation classified.
AI inventory
Every AI tool in use, sanctioned and shadow. SaaS platforms, embedded AI features, browser extensions, code copilots, no-code automation. Categorized by department, use case, and data sensitivity.
Data exposure
What data is flowing into which AI tools. PII, financial records, customer data, IP. Mapped against vendor data-handling commitments and training-data clauses. The assessment identifies the contractual exposure.
Framework gap analysis
Current practices mapped to the NIST AI RMF and, where relevant, to EU AI Act obligations. Control gaps are identified with citations to the framework text. The output is organized for audit review: evidence, control mapping, citations, and named remediation owners. It is not an audit, certification, attestation, or legal advice.
Remediation roadmap
Prioritized actions: which shadow AI to sanction or block, which data flows to restrict, which controls to add. Sequenced by risk reduction per dollar.
For the security seat
What the CISO needs from this
If you are responsible for AI security in this organization, these are the four things this assessment puts in your hands. The same four are what counsel and the auditor may ask you for on their own timeline. Producing them before an incident is usually easier than producing them after one.
A complete AI inventory you can show to an auditor or counsel
Every model, copilot, embedded AI feature, and browser-extension assistant in your environment. Who uses it, what data touches it, what the vendor's training-data and retention terms say. Replaces "we think we have about thirty tools" with a documented list you can defend.
AI data exposure mapped to vendor terms
Which categories of data are flowing into which AI tools: PII, customer records, source code, financials, and IP. Mapped against each vendor's data-use, training, and sub-processor clauses. The output is a vendor-by-vendor exposure register with an entry for each tool.
Documented controls before the incident, not after
Acceptable-use boundaries, DLP coverage on generative AI, identity controls on AI procurement, logging on AI-touching workflows. Each control mapped to the framework it answers (NIST AI RMF and, where relevant, EU AI Act obligations). What "we have controls in place" means, in writing.
A remediation sequence that survives a budget conversation
Prioritized by risk reduction per dollar. Which shadow tools to sanction. Which to block. Which data flows to restrict immediately. Sequenced so the highest-exposure items move first and the work is fundable in pieces, not a single capital ask.
Delivery options
Two ways to engage
Self-service
A short set of questions on AI readiness. It helps you see where you are likely strong and weak, and whether the full AI Governance assessment is worth scheduling. A few minutes; no commitment.
Take the self-assessment →Direct from Preside
Preside delivers the assessment itself. You work with the team that built the method.
Engage Preside directly →Get an answer to the question your board may ask.
AI inventory, data exposure map, framework gap analysis, and remediation roadmap.