AI Governance Framework: How Mid-Market Boards Run One

What a mid-market board should ask for first, and how to build an AI inventory, risk tiers and vendor terms that hold up under Texas law.

By the Total 360 Security team8 min read
Wood-paneled boardroom in a North Texas office tower, with a long table and a closed laptop, where a board reviews its AI gov

An AI governance framework is an operating program for controlling how your company uses AI. It identifies the tools in use, assigns responsibility for their risks, sets data rules, and produces testing records the board can review. Start with the inventory. You cannot govern a tool you cannot name.

For mid-market companies, much of the exposure comes from AI they consume rather than build. An assistant switched on inside existing software can introduce risk without a separate purchasing decision. Your controls sit largely in vendor contracts and account settings, backed by rules employees can actually follow.

What Should the Board Ask For at the November Meeting?

If management cannot identify which commercial AI tools receive company data, the board’s first request should be an inventory with a delivery date. Leave the November meeting with four assigned deliverables:

  1. A complete inventory of AI tools and embedded features, including those switched on by default, with a named owner for each.
  2. A risk tier for each system, based on its use. Any system that shapes decisions about people needs greater scrutiny.
  3. An acceptable-use policy that defines permitted use and tells employees what to do after an accidental disclosure.
  4. A reporting calendar for executives and the audit committee, with consistent items reported each cycle.

A principles statement does not tell directors where company data is going. Assign an executive to own these deliverables and give each one a date.

In the First 30 Days, Build the Inventory Before Writing Policy

In the First 30 Days, Build the Inventory Before Writing Policy, Total 360 Security

Use the first 30 days as an inventory target, not a promise that every unknown will be resolved. Record gaps explicitly. An unanswered question about a vendor’s training practices is a review item, not permission to keep sending data.

  1. Start with expense and corporate card records. Identify paid AI subscriptions, including purchases made outside procurement.
  2. Review software subscriptions for embedded AI features, even where no separate AI charge appears.
  3. Check procurement requests for purchases and trials that were never formally closed.
  4. Review single sign-on logs for connected apps and browser extensions.
  5. Give each entry a business owner. Record its purpose and data classification, then mark it approved or still under review.

Do not buy another AI point tool before you know what is already switched on. The inventory may reveal that your immediate problem is an existing account setting rather than a missing product.

Where Shadow AI Shows Up First

Expense records show paid use before IT necessarily hears about it. They will not reveal every free account. Compare them with sign-on records and admin-console activity rather than treating any one source as a complete inventory.

Browser extensions and connected apps need attention because employees can add AI capabilities without buying a standalone platform. Record the feature employees use, not just the software subscription that contains it.

What Each Inventory Record Must Hold

Record the tool or feature name and its vendor. Add the data it receives and whether the vendor may train on it. Mark anything you cannot verify as unknown and assign someone to resolve it.

Buying an enterprise platform does not eliminate this work. Employees may still use other tools, and the approved platform’s features and settings can change.

How Do You Tier AI Systems by What They Touch?

A tool that screens job applicants deserves more scrutiny than a writing assistant used only with public information. Tier by consequences, not novelty. The following is a working model, not a statutory classification.

TierWhat the system touchesScrutiny and testing
HighShapes decisions about people, communicates publicly, or handles regulated or sensitive dataPre-deployment and periodic testing for accuracy, bias, and misuse; adversarial testing for public-facing systems; documented limits and monitoring
ModerateInternal work involving company or client data, without decisions about people or public outputTesting before rollout and on a risk-based schedule; enforced data rules and a named reviewer
LowProductivity work using public or non-sensitive contentInventory entry and acceptable-use rules; reassessment when the use or data changes

The policy governs employee behavior. The tier determines how closely you scrutinize the system. “Do not put sensitive data into public tools” is insufficient unless employees know what counts as sensitive and which approved alternative they should use.

The policy must also explain what happens after a mistake. If an employee pastes client-confidential or regulated data into an unvetted tool, contain the exposure and preserve the records. Involve counsel and review the vendor’s retention terms before deleting anything.

What Must a Vendor Contract Say About Your Data?

Before signing with an AI vendor, get written answers to these questions. Apply the same review before leaving an embedded AI feature enabled.

  • Are prompts, uploads, or outputs used to train models, including the vendor’s general models?
  • How long is company data retained, and can retention be shortened?
  • Who can access the data, and under what circumstances?
  • Where is the data processed and stored?
  • What happens to the data at termination, and what evidence of deletion is available?
  • How much notice will you receive before the model or its data handling changes?

A sales representative’s verbal assurance is not a control. Have counsel put the required commitments into the contract or data processing addendum. Then check that account settings match those commitments. A negotiated restriction does little operational good if the wrong feature remains enabled.

What Does Texas Law Let You Prove After an AI Incident?

The Texas Responsible Artificial Intelligence Governance Act, known as TRAIGA, has broad reach. Business and Commerce Code Chapter 552 covers people who conduct business in Texas, provide products or services used by Texas residents, or develop or deploy AI systems in Texas. Counsel should determine which provisions apply to your use.

Section 552.105(c) provides a rebuttable presumption of reasonable care for covered entities without conditioning it on certification. Separately, the internal-review discovery route in section 552.105(e)(2) requires substantial compliance with NIST’s Generative AI Profile or another recognized AI risk-management framework. These are specific legal provisions, not a blanket immunity for companies with an AI policy.

NIST published its AI Risk Management Framework 1.0 in January 2023. Its govern, map, measure, and manage functions provide a structure for the program. NIST’s Generative AI Profile, published July 26, 2024, addresses risks specific to generative AI.

Have counsel confirm any claimed protection. Section 552.104(b) addresses timely cure, supported by a statement, evidence, and necessary policy changes. Your incident process needs to preserve the records counsel will need.

Who the Act Reaches

A Dallas-Fort Worth manufacturer should not assume that buying AI rather than developing it puts the company outside TRAIGA. The statute’s scope includes deployment in Texas and products or services used by Texas residents.

Texas state agencies and local governments also have separate mandatory requirements under Government Code Chapter 2054 and DIR rules effective March 18, 2026. A private company should not confuse those public-sector duties with its own obligations.

Which Records Carry the Defense

For each higher-risk system, maintain records that address the investigative-demand categories in section 552.103(b): the current system, its data, metrics, limitations, and monitoring record. Preserve testing evidence and document changes as they happen.

Framework alignment does not replace other applicable duties. Data protection assessments under the Texas Data Privacy and Security Act sit separately at Business and Commerce Code section 541.105.

What Should the Board Receive Each Month and Quarter?

A sound starting cadence is monthly reporting to executives and quarterly reporting to the audit committee. Keep the reporting set consistent so directors can see changes rather than relearn the format.

ItemMonthly to executivesQuarterly to audit committee
Inventory count and changesCurrent count and additionsTrend and material changes
Risk tier changesChanges requiring actionChanges and reasons
Incidents and near missesOpen actions and ownersStatus and corrective action
Vendor reviewsDue and completed reviewsOutstanding issues
Training completionCompletion and gapsCompletion trend
Higher-risk system testingFailures requiring actionResults and unresolved findings

This cadence is an operating recommendation, not a general legal requirement. Increase reporting when a significant incident or deployment warrants executive attention. Every unresolved finding should have an owner and a due date.

Who Runs the Program Month to Month When There Is No AI Officer?

Portrait of a security professional sitting in an office armchair.

You do not need a full-time AI officer to begin. You need one accountable owner with time allocated to maintain the inventory and follow through on findings. A committee can review decisions, but it cannot substitute for ownership.

At Total 360 Security, our fractional Chief AI Security Officer service covers AI inventories, acceptable-use policies, risk assessments, and organizational guardrails. We are Texas DPS-licensed, with CPP, CISSP, and CISM credentials. Our broader security programs cover all 13 risk domains and report directly to the board.

Bring your current tool list and the board’s unanswered questions to a 30-minute risk discussion. Review the scope on our fractional Chief AI Security Officer page before the meeting.

Frequently asked questions

What is an AI governance framework and what does it include?

An AI governance framework defines who owns AI risk and how the organization controls it. It includes an inventory, risk tiers, use policies, testing, incident handling, and documentation aligned with a recognized framework such as NIST’s AI Risk Management Framework. It must stay current as tools and uses change.

Do we need a full-time AI officer or can a fractional Chief AI Security Officer run this?

A fractional Chief AI Security Officer can own the program without a full-time appointment. The engagement needs clear authority, assigned responsibilities, and a reporting calendar. Internal business owners still need to explain their use cases and carry out assigned actions.

What does TRAIGA require of Texas businesses?

TRAIGA carries civil penalties and enforcement provisions, and the duties that apply depend on the organization and its AI use. Its reasonable-care presumption and statutory defenses are not substitutes for compliance. Have counsel assess your obligations and the records needed to support any protection.

How do we find shadow AI tools employees are already using?

Start with expense and corporate card records, then compare them with procurement requests, single sign-on logs, and browser or SaaS admin consoles. Paid subscriptions are only part of the picture. Check embedded features and free tools, and assign an owner to every use you identify.

What vendor contract clauses protect us when a SaaS provider trains on our data?

Have counsel address permitted training use explicitly, including whether company inputs and outputs may train general models. The agreement should also cover retention, access, processing location, termination, and notice of changes. Verify the account settings against the signed terms rather than relying on a sales assurance.

Total 360 Security provides virtual Chief Security Officer, vCISO, and Enterprise Security Risk Management programs for mid-market organizations. They design, run, and report security programs covering all 13 risk domains rather than just offering advisory decks or point solutions. The firm is Texas DPS-licensed with CPP, CISSP, and CISM credentials.

Serving Texas, California, Nevada, New Mexico.