Texas universities turn to AI to police course language on race and gender
Under new rules giving regents more control over curricula, Texas A&M and Texas State are using AI to review course descriptions and syllabi that touch on race, gender, and identity. Administrators frame it as transparency and alignment. Faculty and AI experts see something else: a shift of control away from instructors supported by tools that can't read context.
The stakes are immediate. Texas A&M is testing AI to scan course materials systemwide. Texas State ordered hundreds of faculty to rewrite syllabi, recommending an AI assistant to "neutralize" language by removing words like "dismantling" or "decolonizing."
What's actually happening
At Texas A&M, a senior official used an AI tool (via OpenAI services) to count courses referencing feminism at a regional campus. Slight tweaks in the prompt produced different answers each time. Internally, leaders acknowledged the tool's "inherent risk of inaccuracy."
The system plans to let roughly 20 staff run hundreds of AI queries each semester. New rules also require presidents to sign off on courses that could be seen as advocating "race and gender ideology," and prohibit teaching material not listed on an approved syllabus.
Officials emphasize that people, not software, will make final decisions. Still, early tests showed inconsistent outputs, and the list of search terms used to flag courses has not been shared with faculty.
AI-assisted rewrites at Texas State
Texas State flagged 280 courses for "neutrality" concerns and instructed faculty to revise titles, descriptions, and learning outcomes. Examples include retitling "Combating Racism in Healthcare" to "Race and Public Health in America," and avoiding outcomes like "value diversity" or "embrace activism."
Administrators circulated an AI prompt instructing a chatbot to strip advocacy, prescriptive conclusions, affective outcomes, or ideological commitments-then generate three alternative versions. Faculty faced a tight deadline to keep courses on the spring schedule, creating pressure to comply.
Why experts are uneasy
Large language models don't "understand" content. They predict words that look right based on patterns. That makes them sensitive to prompt phrasing and prone to agreeing with users-what researchers call "sycophancy."
As one linguist put it, these systems are excellent at producing plausible text, not assessing truth or context. A Baylor professor called keyword scanning a "blunt instrument": it can flag a term without any grasp of how or why it's being taught.
Faculty leaders warn this approach sidelines disciplinary expertise and accelerates administrative control over course design. Even supportive voices inside A&M say any tool needs clear standards and real testing before adoption.
What this means for education professionals
If you're in academic affairs, instructional design, or faculty leadership, this isn't a theoretical debate. It's a policy change with operational consequences for catalog governance, academic freedom, and student expectations.
The promise: faster alignment, clearer catalog descriptions, fewer surprises for students. The risk: overreliance on AI pattern-matching that confuses signal with noise, chills discussion, and reduces complex pedagogy to a list of "allowed" words.
Practical guardrails for institutions
- Publish the criteria. Define "advocacy," "neutral," and "affective outcomes" with examples across disciplines. Ambiguity invites arbitrary enforcement.
- Share the search terms. Faculty deserve to see what triggers flags and to contest false positives.
- Require reproducibility. Standardize prompts, temperature, and sampling. Log runs. If results vary with minor phrasing changes, don't use them to make decisions.
- Keep humans in the loop. AI can triage, not judge. Require subject-matter review before any course is paused, retitled, or removed.
- Evaluate context, not keywords. Read the unit, not just the term. Document how a flagged topic supports course outcomes.
- Pilot before policy. Run a limited trial with faculty oversight and publish findings on accuracy, false positives, and workload impact.
- Create an appeals process. Time-bound reviews with faculty representation, plus version control for syllabi and catalog entries.
- Protect academic freedom. Reaffirm that policy enforcement won't punish rigorous, evidence-based instruction of sensitive topics presented within course scope.
- Train the reviewers. Prompt-writing, bias awareness, and AI failure modes should be required training for staff using these systems.
Practical steps for faculty
- Align the basics. Make sure your catalog description, weekly topics, and assessments clearly connect. If you teach sensitive content, state its relevance up front.
- Write measurable outcomes. Focus on knowledge and skills (analyze, compare, evaluate, apply). Avoid language that implies required beliefs or feelings.
- Add context notes. One sentence that explains why a flagged topic supports the course objective can preempt misunderstandings.
- Keep versions. Maintain a simple changelog for your syllabus and catalog text. It speeds reviews and protects your intent.
- Request clarity. If a course is flagged, ask for the exact terms, prompts, and outputs used. Document all responses.
- Use AI carefully. If you try an assistant to edit language, control the prompt, review every change, and reject anything that distorts your pedagogy.
Key tensions to watch
- Policy scope: How broadly do regents apply "advocacy" and "ideology," and who decides?
- Tool reliability: Do outputs remain consistent with fixed prompts and parameters across semesters?
- Faculty voice: Are governance bodies, AI councils, and department chairs meaningfully involved before policies take effect?
- Transparency: Will institutions disclose term lists, prompts, and validation results used in audits?
Bottom line
AI can help surface inconsistencies at scale, but it cannot judge nuance, intent, or pedagogy. Use it as a pointer, never as the arbiter. Clear standards, faculty governance, and reproducible processes matter more than dashboards and keyword hits.
If your campus is testing AI in curriculum audits, start with a pilot, publish the methodology, and treat every AI flag as the start of a human review-not the end of it.
Further resources
- OpenAI services referenced by Texas A&M: openai.com
- Practical prompt writing for academic workflows: Complete AI Training: Prompt Engineering
Your membership also unlocks:
