AI Governance Consulting: Rules for AI Your Staff Will Follow
AI governance consulting sets the rules, owners and controls for the AI tools your staff already use: which tools are approved, what data may go into each, which connectors may reach company mail and files, and who checks the output. NetSys writes those rules with you, sets the technical controls that back them up in Microsoft 365, Google Workspace and the Claude and ChatGPT workspaces themselves, and reviews the program on a schedule.
The short answer
AI governance services give a business the rules, owners and controls that decide which AI tools it may use, with what data, through which connectors, and who is accountable for each use. NetSys builds that program for small and mid-sized firms: an AI acceptable use policy, a register of approved tools with vendor reviews, workspace settings and connector approvals in Microsoft 365 Copilot, ChatGPT Business and Claude Team or Enterprise, device and data loss prevention controls, and scheduled reviews mapped to NIST AI RMF 1.0.
Closest related page: AI security assessment. The assessment is a point-in-time review that produces findings, while governance consulting turns findings like those into a standing program with a policy, owners, controls and scheduled reviews.
Governance matters most where AI touches other people's information: law and accounting firms, medical and dental practices, financial advisers, and any business whose clients send security questionnaires. It also matters if an AI tool screens candidates for New York City jobs, or if an AI tool's output is used in the European Union, because specific rules apply there.
A small firm does not need an AI committee. It needs one named owner, a short policy people actually read, a list of approved tools on business accounts, and controls that make the safe path the easy one.
How is the program built? Inventory, policy, controls, then review
We begin with what is already happening: sign-in and app-consent logs, managed browsers and a short staff survey show which AI tools are in use, on which accounts. The policy is drafted with your managers and reviewed by your counsel, and each approved tool gets an owner, an account type and a data rule. Controls start in audit-only mode, so you see what they would block before they block it. After that, the register, logs and policy are reviewed on a set schedule.
What Does AI Governance Consulting Cover?
AI Acceptable Use Policy
Short enough to read, specific enough to follow.
- Which data types may go into which approved tools, and which never may
- Rules for client, patient, employee and financial information
- When AI-drafted work must be checked by a person before it leaves the firm
- How staff request a new tool, and who approves it
AI Inventory and Use-Case Register
You cannot govern tools nobody has found.
- AI tools and embedded AI features found through sign-in logs, app consents and a staff survey
- An owner, an account type and a risk level recorded for every approved use
- Vendor terms checked for retention, model training and subprocessors
- Unapproved tools replaced with an approved option or retired
Controls That Back the Policy
The approved path should also be the easy one.
- Business AI accounts for approved tools, tied to company sign-in
- Warnings or blocks on unapproved AI sites from managed devices
- Data loss prevention rules on pasting sensitive content into restricted sites, where licensing allows
- Admin approval before staff can connect AI apps to mailboxes and files
- Workspace settings in ChatGPT Business and Claude Team or Enterprise: single sign-on, who may join, which apps and connectors are on, and who may publish custom MCP connections
- Microsoft 365 sharing cleaned up before Copilot can surface it
Frameworks, Laws and Reviews
Mapped to the standards clients and insurers ask about.
- Program mapped to the NIST AI RMF functions: Govern, Map, Measure and Manage
- An ISO/IEC 42001-style management structure for firms that want one
- NYC Local Law 144 and EU AI Act questions flagged for counsel where they apply
- Scheduled reviews of the register, the logs and the policy
Why do firms choose NetSys for AI governance?
Tell us which AI tools your team uses, what client data you hold and who is asking about AI. We will outline the policy, controls and review cadence that fit a firm your size.
Who does the work: meet the NetSys team on our About page.
- The policy and the controls come from the same team, so the rules are enforced as well as written
- A managed IT and cybersecurity firm since 1998, so AI is governed like every other system we run
- Microsoft 365 and Google Workspace administration handled in-house
- Honest scope: we build and evidence the program, and your counsel makes the legal calls
- Month-to-month terms for the ongoing reviews
Measured results from our case studies
AI Governance Consulting in practice
- Importers and distributors
Regional food importing company
Microsoft Copilot and Claude rollout for 25 employees, AI usage policies, connector access controls, and staff training.
Microsoft Copilot ConsultingIT for Importers and Distributors
Client names are withheld. Each card is the scope of a real NetSys engagement, as delivered.
Scope, process and planning details
Scope, options and working examples
What AI governance consulting includes, and what it does not
Governance fails when a written rule is mistaken for a working control, so the scope keeps the two apart.
| Area | Included | Not included |
|---|---|---|
| Policy | An AI acceptable use policy, data rules by type and an approved-tool list | Legal sign-off, which stays with your counsel |
| Inventory | Discovery from sign-in logs, app consents, managed browsers and a staff survey | A guarantee that every past use of AI has been found |
| Controls | Configuration in your Microsoft 365 or Google Workspace tenant and on managed devices | Monitoring personal phones or home computers |
| AI workspaces | Settings, connector approvals and sharing rules in ChatGPT Business, Claude Team or Enterprise and Microsoft 365 Copilot | Building assistants, agents or integrations, which are scoped as separate projects |
| Frameworks | Mapping to NIST AI RMF 1.0, and an ISO/IEC 42001-style structure on request | An ISO/IEC 42001 certification audit |
| Hiring tools | Identifying AI in screening software and what NYC Local Law 144 asks of you | The independent bias audit the law requires |
| Reviews | Scheduled reviews of the register, the logs and the policy | Reading every prompt your staff write |
Start from the staff-facing rules: Acceptable use policy template with an AI tools section.
This is general information, not legal advice. Where a law such as NYC Local Law 144 or the EU AI Act may apply, we flag it for your counsel.
Who owns what in an AI governance program
Accountability is what turns a policy into a program. Each role is named in writing at kickoff:
| Role | Usually held by | Accountable for |
|---|---|---|
| Executive owner | The owner, a partner or a senior manager | Approving the policy, accepting or refusing risks, and settling disputes over a tool |
| Tool owner | The manager of the team that uses the tool | The approved use, its data rule and who holds an account |
| IT and security | Your IT lead, NetSys, or both | The inventory, the controls, the logs and the scheduled reviews |
| Legal counsel | Your in-house or outside counsel | Legal calls, such as NYC Local Law 144 or EU AI Act exposure |
| Every employee | Everyone with an approved AI account | Following the policy, checking AI output before it leaves the firm and reporting mistakes |
A small firm does not need an AI committee. One person often holds more than one role, as long as each role has a name against it.
What an AI vendor review checks
Every tool on the approved list has its terms read before staff get an account. The review asks:
- Does the vendor train its models on your content, and does the business plan switch that off?
- How long are prompts, files and outputs kept, and can you delete them?
- Which subprocessors handle your data, and where?
- Do accounts sign in through your company identity, so access ends when someone leaves?
- Which admin controls and activity logs come with the business plan?
- Which connectors can reach mail, files or other systems, and who approves them?
The answers are kept with the tool's entry in the register and checked again at each scheduled review.
How do you govern Claude use, connectors and data?
For a firm on Claude Team or Claude Enterprise, governance comes down to six decisions. Each is recorded in the register and enforced in a specific setting, named here as Anthropic's support documentation describes it as of October 2026:
| Decision | What to decide | Where it is enforced |
|---|---|---|
| Workspace and seats | Company work moves off personal Pro and Max accounts onto Claude Team (2 to 150 seats) or Claude Enterprise (20-seat minimum, with SCIM, audit logs and custom data retention) | Organization settings; SSO and domain capture on both plans; personal accounts merged or closed through Anthropic's migration flow |
| Approved data | Which client, patient, employee and financial data may go into Claude; commercial plans do not use inputs or outputs for training by default, and deleted chats leave back-end storage within 30 days | The policy, role-based training and data loss prevention on managed devices |
| Connectors and MCP servers | Which connectors the firm allows; on Team only Owners and Primary Owners can add custom connectors, then each member connects individually; the Microsoft 365 connector needs a Microsoft Entra Global Administrator's consent and is read-only by default | Organization settings > Connectors; Entra admin consent; write tools enabled per tool by an owner |
| Projects and Skills | Who may create and share Skills and Projects; owners can turn off user-created skills and set library publishing to Requires review | Organization settings > Plugins & skills; the project sharing policy |
| Cowork and Claude Code | Whether cloud sessions run, what network access they get, whether the Auto approval mode is allowed, and which MCP servers Claude Code may use | Organization settings > Cowork; server-managed settings for Claude Code |
| Human review and incidents | Which outputs a person must check before they leave the firm, which agent actions always wait for approval, and how staff report a mistake | The policy, Project instructions and approval steps in each assistant or agent |
Supporting guides for the Claude decisions: How to review a Claude MCP connector before staff use it · Claude Microsoft 365 connectors: permissions before rollout · Claude consulting and business implementation · AI security assessment of a Claude deployment.
Setup steps stay in the supporting guides; this page is the policy and the review. Owners can also disable interactive tool calls per connector, and Anthropic reviews directory connectors against its listing criteria but does not security-audit MCP servers, so the firm's own review matters.
What should a ChatGPT Business workspace policy cover?
ChatGPT Business (formerly ChatGPT Team) ships with defaults that suit a small team trying it out, not a firm holding client data. The policy records each decision and the setting that enforces it, as OpenAI documents them as of October 2026:
| Decision | What to decide | Where it is enforced |
|---|---|---|
| Accounts and sign-in | Company work moves off personal Plus and Pro accounts into a Business workspace (2 to 200 seats) or ChatGPT Enterprise; SAML or OIDC single sign-on and a verified domain are included on Business, SCIM is Enterprise only, and multifactor authentication is enforced through your identity provider because ChatGPT admins cannot require it | Workspace settings > Identity & access; the MFA or Conditional Access policy in Entra ID or Google Workspace |
| Who can join and invite | Workspace discovery is on by default on Business, so anyone with a verified-domain email can find and ask to join, and members can invite other members, which cannot be restricted, so the policy says who may | Identity & access > User Provisioning; the owner role and the policy |
| Approved data and retention | Which data types may go into ChatGPT; OpenAI does not train on Business or Enterprise workspace data by default, admins control how long data is retained, deleted conversations leave OpenAI's systems within 30 days, and no BAA is offered on Business, so regulated health data needs a sales-led plan | Workspace retention settings; data loss prevention on managed devices; the policy |
| Apps, plugins and custom MCP servers | Apps are enabled by default on Business, so the owner decides which stay on; each app's approval level runs from Always ask to Allow all actions; only admins and owners can publish plugins for custom MCP servers; Microsoft apps need a Microsoft Entra administrator's consent | Admin Console > Plugins and the Apps settings; Entra admin consent |
| Custom GPTs and workspace agents | Who may build and share GPTs and workspace agents, and with which connections; custom GPTs are scheduled to retire on December 11, 2026 in favor of plugins, so new builds go to plugins or workspace agents | Workspace settings > GPTs; agent building and publishing permissions |
| Memory, sharing and spend | Whether Memory is on for the workspace (owners can turn it off), the fact that shared links stay inside the workspace but cannot be disabled on Business, and monthly credit limits per seat type, which default to none | Workspace settings; Billing > spend controls |
Supporting guides for the ChatGPT decisions: ChatGPT Business security settings and data privacy checklist · A one-page AI usage policy to copy · Shadow AI: when employee ChatGPT use leaks data · ChatGPT consulting and business workspace setup.
ChatGPT Enterprise adds the controls a regulated firm may need: SCIM, role-based access, the Compliance Platform that feeds Microsoft Purview and other data loss prevention tools, IP allowlisting and data residency. Setup steps live in our ChatGPT Business security settings checklist.
What the governance assessment delivers
The first phase ends with documents you keep and controls already running:
- An inventory of the AI tools and embedded AI features in use, and the accounts they run on
- An AI acceptable use policy, drafted with your managers and reviewed by your counsel
- A register of approved tools and uses, each with an owner, an account type, a risk level and a data rule
- Vendor review notes for every approved tool
- Control settings documented, with what the audit-only period showed before anything was blocked
- A review calendar, starting with the first scheduled check of the register, logs and policy
Each document is mapped to the NIST AI RMF functions: Govern, Map, Measure and Manage.
How an AI governance engagement starts
Most programs reach a signed policy and working controls in this order.
- Kickoff: an executive owner named, and the teams, tools and data types in scope agreed
- Discovery: AI use found through logs, managed browsers and a staff survey
- Policy: drafted with your managers, reviewed by counsel, then taught in a short session per role
- Controls: approved accounts issued, then blocking and data rules switched on after an audit-only period
- Review: the first scheduled check of the register, logs and policy, with changes recorded
Pricing and scope
How AI governance consulting is priced
There is no price list. The quote follows the kickoff call and moves with these factors.
- Headcount and the number of teams the policy covers
- How many AI tools and embedded AI features are in use
- Your Microsoft 365 or Google Workspace licensing, which decides which controls are available
- Regulatory scope, such as HIPAA, SEC Regulation S-P, NYDFS 23 NYCRR 500, NYC Local Law 144 or EU exposure
- A one-time program build, or ongoing reviews under a month-to-month agreement
For managed clients, ongoing governance reviews can be added to the per-user monthly agreement.
Guides and checklists
Related guides and tools
- AI security assessment of your current tools
- How to review a Claude MCP connector before staff use it
- Claude Microsoft 365 connectors: permissions before rollout
- Claude consulting and business implementation
- ChatGPT Business workspace setup with the policy applied
- ChatGPT Business security settings and data privacy checklist
- AI and ChatGPT training for employees, by role
- Managed AI services and ongoing workspace administration
- An AI governance framework sized for a small business
- AI governance tools for small businesses, compared
- AI usage policy template guide
- Stopping shadow AI data leaks
- AI data privacy questions small business owners ask
- Fixing SharePoint oversharing before Copilot
- Data loss prevention for Microsoft 365 and endpoints
- What shadow AI means
- AI consulting to choose the right tools first
- Draft a policy for staff use of AI
AI Governance Consulting FAQs
What is AI governance?
AI governance is how an organization decides which AI tools may be used, for which tasks and with which data, and who is accountable for each use. In practice it is a written policy, a register of approved uses with named owners, technical controls that enforce the rules and scheduled reviews. Frameworks such as NIST AI RMF 1.0 and ISO/IEC 42001 give the program a structure.
What do AI governance consulting services include?
A NetSys governance engagement delivers an inventory of the AI tools in use and the accounts they run on, an acceptable use policy drafted with your managers and reviewed by your counsel, a register of approved tools and uses with an owner and a data rule for each, vendor review notes, workspace and connector settings applied in Microsoft 365 Copilot, ChatGPT Business and Claude, device controls such as data loss prevention where licensing allows, a short training session per role and a review calendar. NetSys works from Brooklyn, on site across the New York area and remotely for firms elsewhere in the US.
What is the role of AI governance?
Its role is to let a business use AI without losing control of three things: which data reaches AI tools, whether AI output is checked before anyone relies on it, and who answers for each use. It also gives staff a clear yes on which tools they may use, which makes the approved path easier than the unapproved one.
Who is responsible for AI governance in a small business?
One named executive owner, usually the owner or a partner, approves the policy and makes the risk calls. Each approved tool gets an owner from the team that uses it, IT runs the controls and the reviews, and counsel makes the legal calls. A small firm needs those roles in writing, not an AI committee.
How does AI governance relate to responsible AI and ethics?
Governance is how ethical principles get enforced. NIST AI RMF 1.0 describes trustworthy AI as valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. In a small firm those become concrete rules: a person checks AI output before it reaches a client, hiring tools are reviewed where NYC Local Law 144 applies, and client data stays out of tools that would train on it.
Does a small business need an AI acceptable use policy?
Yes, if anyone on staff uses AI tools, and in most firms someone already does. Without a written policy, each employee decides alone what client or employee data goes into which tool. A short policy naming approved tools, banned data types and a review step for client-facing output protects the business and gives staff a clear yes.
How do you stop staff pasting client data into ChatGPT?
Give staff an approved business AI account, such as ChatGPT Business or Claude Team on company sign-in, and make unapproved tools harder to use than the approved one. On managed devices we can warn or block on unapproved AI sites, and data loss prevention rules can audit or block pasting sensitive content into restricted sites where your Microsoft licensing supports it. Policy and short training cover the rest.
How do you govern Claude use, connectors and data?
Move company work onto Claude Team or Enterprise so an owner controls seats, single sign-on and settings; write down which data types may go into Claude; let only owners add connectors and keep the Microsoft 365 connector read-only until a named owner approves specific write tools; decide who may create and share Skills and Projects; set the Cowork and Claude Code policies for cloud sessions, network access and approvals; and name the outputs a person must check. Anthropic's commercial terms do not use your inputs or outputs for training by default, as of October 2026, but the connector and sharing decisions are yours to make and review.
More frequently asked questions
What should a ChatGPT Business workspace policy cover?
Five things, each tied to a setting. Accounts: company work happens in the Business workspace through single sign-on and a verified domain, not in personal accounts. Data: which data types may be entered, knowing that OpenAI does not train on Business workspace data by default and that no BAA is offered on Business. Apps: which connected apps stay enabled (they are on by default), what approval level each gets and who may publish custom MCP plugins (only admins and owners). Builders: who may create and share GPTs, plugins and workspace agents, with custom GPTs scheduled to retire on December 11, 2026. Housekeeping: memory, shared links, spend limits and how often the owner reviews members and apps. All as OpenAI documents it in October 2026.
What should an AI governance policy cover?
An AI governance policy should cover approved tools, the data allowed in each, human review before AI output reaches a client, and an owner for each use. It should also set how staff request new tools, how AI mistakes are reported, how vendors are vetted and how often the policy is reviewed, in few enough pages that people read it.
Does AI policy consulting include employee training?
Yes. A policy nobody has been taught is a document, not a control. Each role gets a short session on the approved tools, the data rules, how to check AI output and how to report a mistake, run in the tools people actually use. Deeper hands-on training by role is a separate service, described on our AI training page, and the policy session can be folded into it.
Which AI governance frameworks do you use?
We map programs to the NIST AI Risk Management Framework (AI RMF 1.0), a voluntary framework built on four functions: Govern, Map, Measure and Manage, and add its Generative AI Profile, NIST AI 600-1. Firms that want more structure can align to ISO/IEC 42001:2023; certification under it comes only from an accredited certification body, not from NetSys.
Does NYC Local Law 144 apply to our hiring tools?
It may, if an AI or machine-learning tool substantially assists hiring or promotion decisions for jobs tied to a New York City office. The law then requires an independent bias audit within the past year, a published summary of the results and notice to candidates who live in the city. This is general information, not legal advice.
Does the EU AI Act apply to a US company?
The EU AI Act can apply to a US company: it covers providers and deployers outside the EU when an AI system's output is used in the EU. Prohibited practices have applied since February 2, 2025; after the 2026 AI Omnibus amendment, high-risk rules for uses such as hiring apply from December 2, 2027. General information, not legal advice.
Sources and technical references
- NIST: AI Risk Management Framework (AI RMF 1.0, January 2023)
- NIST AI RMF 1.0, section 3: characteristics of trustworthy AI
- NIST AI 600-1: Generative AI Profile (July 2024)
- IEC: ISO/IEC 42001:2023, AI management system
- NYC DCWP: automated employment decision tools (Local Law 144 of 2021)
- NYC DCWP: Local Law 144 frequently asked questions
- EUR-Lex: Regulation (EU) 2024/1689, the EU AI Act
- European Commission: AI Act timeline after the 2026 AI Omnibus amendment
- Microsoft Learn: Endpoint data loss prevention
- Microsoft Learn: how users consent to applications in Entra ID
- Anthropic: custom connectors (remote MCP) and who can add them
- Anthropic: Microsoft 365 connector setup, admin consent and write tools
- Anthropic: provisioning and managing Skills for an organization
- Anthropic: Claude Cowork architecture and admin controls
- Anthropic: commercial product data is not used for training by default
- Anthropic: how long organization data is stored
- Anthropic: Claude Team and Enterprise plans, seat ranges and features
- OpenAI: what ChatGPT Business is, seats and training defaults
- OpenAI: billing and the 200-seat limit in ChatGPT Business
- OpenAI: members, roles, invitations and workspace discovery in ChatGPT Business
- OpenAI: custom GPT retirement on December 11, 2026 and migration to plugins
- OpenAI: data sharing, shared links and privacy in ChatGPT Business
- OpenAI: BAA availability (none for ChatGPT Business)
- OpenAI: Memory controls in ChatGPT Business
- OpenAI: spend controls and credit limits in ChatGPT Business
- OpenAI: SSO and domain verification for ChatGPT Business
- OpenAI: app, plugin and approval controls for Business and Enterprise workspaces
- OpenAI: enterprise privacy, training defaults and retention
Name an owner, then write rules you can enforce.
Tell us which AI tools your staff use and what data your business holds. We will review your AI usage policy, or outline one, with the controls behind it and a review cadence sized to your firm.