Quick Overview
- This standard sets the specific requirements for how SFA governs the buying, building, and use of artificial intelligence (AI), including generative AI tools.
- It carries out the goals in the SFA 06-107.1 Information Security Organization, Personnel & Privacy Policy by extending SFA's risk-based approach to AI.
- It affects everyone who uses AI for University work: use only approved tools, match the tool to your data classification, and stay accountable for AI outputs.
- People must stay in charge of consequential decisions, and SFA must tell individuals when they are interacting with AI.
- These requirements are the minimum baseline SFA must meet; the University may choose to do more.
- Meeting this standard is mandatory unless a formal exception is granted.
Artificial intelligence, especially generative AI, can help SFA work faster and smarter, but it also creates new risks to privacy,
security, fairness, and accuracy. This standard spells out what the University must do to adopt AI in a way that is lawful, secure,
ethical, transparent, and accountable. It covers the whole AI lifecycle, from deciding to buy or build a tool through design, testing,
everyday use, monitoring, and retirement. It also protects the people affected by AI, especially when AI helps make decisions about them.
|
i
|
Policy vs. Standard
The matching policy (06-107.1) says what SFA wants to achieve. This standard says how, the concrete requirements for AI. Day-to-day steps will live in procedures and the SFA AI Use Guidelines.
|
Who This Applies To
This standard applies to all SFA employees, students, contractors, consultants, and authorized users who procure,
develop, configure, deploy, integrate, or use AI systems for institutional purposes. It covers every kind of AI, whether the tool is
University-managed, cloud-hosted, embedded inside another product, or accessed as a free public service. It applies with extra care to
AI that informs or makes consequential decisions about individuals (such as admissions, financial aid, grading, employment,
benefits, discipline, or access to University services). It does not override academic freedom, instructor course policies, or approved
research protocols, but those uses still must follow the data, prohibited-use, and disclosure rules here.
|
i
|
Baseline, not a ceiling
Everything here is the minimum SFA must do. Departments and instructors may add stricter AI requirements, but never anything weaker. Where rules conflict, follow the more restrictive one.
|
Who Is Responsible
Key roles & responsibilities
⌄
A note on names: At SFA, the senior information security role is the Chief Information Security Officer (CISO). There is no separate "Information Security Officer" position. The abbreviation ISO at SFA refers to the Office of Information Security (the office that supports the CISO), not to a person.
- SFA Chief Information Security Officer (CISO): Owns this standard and SFA's AI governance program. Sets the AI risk strategy, leads (or names the leader of) the AI Governance Committee, approves the AI inventory and risk-tiering method, and authorizes high-risk and consequential-decision AI before it goes live.
- AI Risk Officer: The role accountable for coordinating AI risk decisions across the University. At SFA, the AI Risk Officer role is fulfilled by the CISO.
- Privacy Officer (PO): Makes sure AI use follows privacy law and policy (FERPA, HIPAA where it applies, the Texas Data Privacy and Security Act, and TRAIGA), and reviews AI that handles PII or makes decisions about individuals.
- Data Management Officer (DMO): Confirms the data classification of information proposed for AI use and authorizes (or denies) which data may be entered into each approved AI tool.
- AI Governance Committee: A cross-functional body that reviews AI intake requests and risk classifications, oversees the AI inventory, reviews bias and monitoring results, approves exceptions, and escalates high-risk matters.
- Users / Deployers of AI: Everyone who uses or sets up AI for University work. They use only approved tools for the right data classification, follow acceptable-use and prohibited-use rules, give required disclosures, verify AI outputs, and report AI problems.
Full role definitions
Complete responsibilities for each role are in the attached standard PDF.
What This Standard Requires
The requirements are grouped into seven areas. Expand any area for a plain-language summary of what it requires.
The official requirement numbers (like 06-107.1.8.1.1) are shown so you can match them to the attached PDF.
1 AI Governance & Accountability
NIST AI RMF
The framework & who owns it
⌄
SFA must have a written framework for governing AI, plus a committee, clear ownership, and training.
- Establish and maintain a documented AI governance framework built around the NIST AI RMF functions (GOVERN, MAP, MEASURE, MANAGE) and the trustworthy-AI characteristics (06-107.1.8.1.1).
- Stand up an AI Governance Committee with a charter and decision authority, and identify a responsible owner for every AI system in use (06-107.1.8.1.2).
- Define the University's AI risk tolerance and strategy, and fold AI risk into SFA's overall enterprise and information security risk management (06-107.1.8.1.3).
- Align AI rules with SFA 06-107.1, the Data Protection Standard, the SFA AI Use Guidelines, UTS 165, and Texas law, resolving conflicts in favor of the stricter rule (06-107.1.8.1.4).
- Provide AI awareness and role-based training; completing DIR statewide AI awareness training meets the baseline for covered personnel (06-107.1.8.1.5).
2 AI System Inventory & Risk Classification
Knowing what we have & how risky it is
⌄
SFA must track its AI systems and sort them by risk, giving extra scrutiny to anything that affects people.
- Keep a current, authoritative inventory of AI systems, including embedded AI features, recording owner, purpose, data handled, vendor, and risk tier (06-107.1.8.2.1).
- Use a documented risk-tiering method (for example minimal, limited, high, and prohibited) and classify each system before deployment (06-107.1.8.2.2).
- Identify AI that makes or informs consequential decisions about individuals as high-risk, requiring human oversight, bias testing, and CISO authorization (06-107.1.8.2.3).
- For each high-risk system, complete and retain a MAP-phase impact analysis of purpose, stakeholders, benefits, harms, and controls (06-107.1.8.2.4).
3 AI Procurement & Third-Party AI
Vetting tools before we buy or use them
⌄
AI tools must be assessed before purchase or use, and contracts must protect University data.
- Complete a University AI risk assessment, coordinated by the Office of Information Security, before any AI product or feature is bought or used (06-107.1.8.3.1).
- Require contract terms that bar vendors from training on University data without permission and that cover ownership, retention, secure deletion, breach notice, and TRAIGA compliance (06-107.1.8.3.2).
- Evaluate value-chain and component risk, requiring vendor documentation such as model/system cards, HECVAT, SOC 2, and TX-RAMP status (06-107.1.8.3.3).
- Reserve final deploy authorization to the CISO; do not use free or click-through AI services with University data above the Published/Public classification unless expressly authorized (06-107.1.8.3.4).
4 Data Input & Output Controls by Classification
What data may go into which tool
⌄
The data you enter into an AI tool is governed by its classification. Match the data to a tool approved for it.
- Only Published/Public data may go into general-purpose, unmanaged, or free AI tools; Controlled and Confidential data (PII, FERPA records, health data) requires a University-approved tool authorized for that classification (06-107.1.8.4.1).
- Use Confidential or Controlled data only with tools configured to block model training on it, isolate it from others, and disable web features; the DMO confirms each tool's authorized classification and users must not exceed it (06-107.1.8.4.2).
- Treat AI outputs as carrying the highest classification of any data used to create them, and do not republish outputs from Confidential or Controlled inputs without review (06-107.1.8.4.3).
- Define retention and disposal for AI prompts, outputs, logs, and derived data, and retain records subject to the Texas Public Information Act and University retention schedules (06-107.1.8.4.4).
5 Acceptable Use & Prohibited Uses
TRAIGA
What you may and may not do
⌄
Use AI only for legitimate University purposes, and remember: you remain accountable for AI-assisted work as if you produced it yourself. Some uses are flatly prohibited.
- Use AI only for legitimate institutional purposes and verify outputs for accuracy and compliance before relying on or sharing them (06-107.1.8.5.1).
- Follow the TRAIGA prohibited-use limits described below (06-107.1.8.5.2).
- Never use AI to produce unlawful sexual content, including material depicting minors or non-consenting individuals, or to deceive in violation of law; report any such use as an incident (06-107.1.8.5.3).
- Follow academic-integrity policy and instructor rules for coursework; AI detection tools may not be the sole basis for a misconduct finding (06-107.1.8.5.4).
- Route new or non-standard AI uses through the AI intake process; exceptions require documented, time-limited CISO risk acceptance (06-107.1.8.5.5).
|
!
|
Prohibited AI uses (TRAIGA)
The University prohibits using AI to:
- Incite or encourage self-harm, harm to others, or criminal activity.
- Create a "social score" that leads to detrimental or unfavorable treatment based on personal behavior or characteristics.
- Uniquely identify individuals from biometric data or images gathered without consent in a way that infringes legal or constitutional rights.
- Infringe, restrict, or impair constitutional rights.
- Unlawfully discriminate against protected classes.
- Produce unlawful sexual content or content intended to deceive in violation of law.
|
6 Transparency, Disclosure & Human Oversight
TRAIGA
Telling people & keeping humans in charge
⌄
People must be told when they are dealing with AI, and a qualified human must stay responsible for decisions that affect individuals.
- Clearly and conspicuously disclose, in plain language and without deceptive design, when an individual is interacting with an AI system such as a chatbot (06-107.1.8.6.1).
- Never let AI make a consequential decision alone; a qualified human must meaningfully review it, be able to override it, and remain accountable for the final decision (06-107.1.8.6.2).
- Keep AI systems transparent enough about purpose and limits that people can understand and contest or seek review of AI-informed decisions (06-107.1.8.6.3).
- Review AI-generated content used in official University communications for accuracy, label it where needed, and keep authorship accountability with the responsible employee or unit (06-107.1.8.6.4).
7 Bias, Fairness, Testing & Monitoring
NIST AI RMF
Testing, watching & responding
⌄
High-risk AI must be tested before use, watched while running, and shut down or fixed when it causes harm.
- Test high-risk and consequential-decision AI before deployment for validity, reliability, security, and harmful bias across demographic groups, recording methods and results (06-107.1.8.7.1).
- Continuously monitor production AI for drift, degraded accuracy, emergent bias, security threats (such as prompt injection and data poisoning), and misuse (06-107.1.8.7.2).
- Report AI incidents and harms to the Office of Information Security and handle them through incident response, escalating to the PO, General Counsel, DIR, and UT System as required (06-107.1.8.7.3).
- Establish and test procedures to contain, roll back, disable, or decommission harmful or out-of-bounds AI, with secure data disposal (06-107.1.8.7.4).
- Feed testing, monitoring, and incident findings back into the governance framework, with CISO review of the overall AI program at least annually (06-107.1.8.7.5).
Conformance & Exceptions
Meeting this standard is mandatory for everyone who procures, develops, deploys, or uses AI for University purposes, and it
takes effect upon adoption. Business units and system owners must implement these requirements for the AI they control and keep evidence of
compliance (inventory entries, risk classifications, authorizations, testing and monitoring records, and disclosures). If a requirement genuinely
can't be met, submit a request through the AI intake process; exceptions require documented, time-limited risk acceptance by the
SFA Chief Information Security Officer (CISO).
|
!
|
Important
Violations may lead to suspension of AI access and disciplinary action under University policy. Uses that violate the Texas Responsible Artificial Intelligence Governance Act (TRAIGA) are subject to exclusive enforcement by the Texas Attorney General, which includes a 60-day cure period and civil penalties.
|
Compliance Mapping (Reference)
For auditors and security staff. Everyday users can skip this section.
Frameworks & control references
⌄
Each requirement in this standard maps to one or more of these authoritative sources:
- NIST AI RMF 1.0, the GOVERN, MAP, MEASURE, and MANAGE functions.
- NIST Generative AI Profile (NIST AI 600-1), for generative-AI-specific risks and actions.
- Texas Responsible Artificial Intelligence Governance Act (TRAIGA / HB 149), Tex. Bus. & Com. Code Ch. 552, effective January 1, 2026.
- UT System UTS 165 AI Governance provisions.
- NIST 800-53 Rev 5, for example the PM, RA, SA, SR, SI, PT, and AT control families.
- TAC 202, for example 202.71, 202.73, and 202.74.
The full requirement-by-requirement control mapping is in the attached standard PDF.
Related Policies & Standards
Supporting policy: SFA 06-107.1 Information Security Organization, Personnel & Privacy Policy
Related standards:
Other references:
- SFA AI Use Guidelines
- NIST AI RMF 1.0
- NIST Generative AI Profile (AI 600-1)
- Texas Responsible Artificial Intelligence Governance Act (HB 149)
- UT System UTS 165
Responsible office: Office of Information Security ·
Contact: itsecurity@sfasu.edu
|
📎
|
Official document
The complete, official standard is attached to this article as a PDF, including its full requirement text, role definitions, and control mappings. This article summarizes that standard in plain language. If anything in this article conflicts with the attached PDF, the PDF is the official version and takes precedence.
|
Need Help?
Contact the IT Help Desk at
(936) 468-4357 (HELP) or submit a ticket at
help.sfasu.edu.
For questions about this standard, contact the Office of Information Security at
itsecurity@sfasu.edu.