Sign up for The Brief, a daily newsletter from The Texas Tribune that keeps readers informed on the most important Texas news.
This fall, a senior official from the Texas A&M University System began testing a new artificial intelligence tool, inquiring about the number of courses discussing feminism at one of its regional institutions. Each query, phrased differently, yielded varying results.
“Either the tool is learning from my previous queries,” Korry Castillo, the Texas A&M System’s chief strategy officer, communicated to colleagues via email, “or we need to refine our requests to achieve the most accurate results.”
On September 25, Castillo was working to fulfill a commitment made by Chancellor Glenn Hegar and the Board of Regents: to conduct an audit of courses across the system’s twelve universities. This initiative followed intense conservative backlash regarding a gender-identity lesson at the flagship campus earlier that month, which resulted in a professor’s dismissal and the university president’s resignation.
Texas A&M officials indicated that the controversy arose from discrepancies between the course content and its description in the university catalog, framing the audit as a means to ensure students receive clear information about what they are enrolling in. As similar scrutiny fell upon other public universities, many began preparing for compliance with a new state law that enhances the authority of governor-appointed regents over curricula, prompting them to announce audits as well.
Records acquired by The Texas Tribune provide an initial glimpse into how Texas universities are leveraging AI to facilitate these reviews.
At Texas A&M, internal emails reveal that staff members are utilizing AI software to scan syllabi and course descriptions for language that might raise concerns under new policies governing how faculty address race and gender issues.
Meanwhile, at Texas State University, memos indicate that administrators are encouraging faculty to employ an AI writing assistant to revise course descriptions. They have advised professors to eliminate terms such as “challenging,” “dismantling,” and “decolonizing,” and to rename courses like “Combating Racism in Healthcare” to more neutral titles such as “Race and Public Health in America.”
While university officials characterize these initiatives as innovative strategies promoting transparency and accountability, AI experts caution that such systems do not genuinely analyze or comprehend course content. Instead, they generate responses that appear correct based on patterns from their training data.
This means that slight variations in question phrasing can lead to disparate results, rendering these systems unreliable for determining whether a course aligns with its official description. Experts warn that using AI in this manner could result in courses being flagged over isolated terms, further shifting control of education from faculty to administrators.
“I’m not convinced this is about serving students or improving syllabi,” remarked Chris Gilliard, co-director of the Critical Internet Studies Institute. “This appears more like a project aimed at controlling education and transferring it from professors to administrators and lawmakers.”
During a recent board of regents meeting, Texas A&M System leaders outlined the new processes being developed for course audits as a systematic enforcement mechanism.
Vice Chancellor for Academic Affairs James Hallmark mentioned that the system would utilize “AI-assisted tools” to analyze course data based on “consistent, evidence-based criteria,” which would inform future board decisions regarding courses. Regent Sam Torn lauded this initiative as “real governance,” asserting that Texas A&M was “leading the way and setting the standard for others to follow.”
On the same day, the board enacted new regulations mandating that presidents approve any course perceived to advocate for “race and gender ideology,” prohibiting professors from teaching materials not included in the approved syllabus.
In a statement, Chris Bryan, the system’s vice chancellor for marketing and communications, explained that Texas A&M is utilizing OpenAI services through an existing subscription to assist with the course audit. He noted that the tool is still in the testing phase as universities finalize their course data. Bryan emphasized that decisions regarding appropriateness, alignment with degree programs, or student outcomes will ultimately be made by individuals, not software.
Records indicate that Castillo informed colleagues to expect around 20 system employees to utilize the tool for hundreds of queries each semester.
The documents also reveal some concerns that emerged during initial tests of the tool.
When Castillo shared her experience of obtaining inconsistent results while searching for classes related to feminism, deputy chief information officer Mark Schultz warned that the tool carries an “inherent risk of inaccuracy.”
“Some of that risk can be mitigated through training,” he noted, “but it likely cannot be completely eliminated.”
Schultz did not elaborate on the specific kinds of inaccuracies involved. When asked whether the potential inaccuracies had been addressed, Bryan stated, “We are currently validating the accuracy, relevance, and repeatability of the AI tool’s responses through baseline conversations.” He explained that this involves assessing how the tool reacts to misleading prompts and ensuring human oversight in reviewing the results.
Experts indicated that the differing responses Castillo received upon rephrasing her questions illustrate the operational nature of these systems. They explained that such AI tools generate responses by predicting patterns and constructing text strings.
“These systems fundamentally function by answering the question ‘what is the likely next word,’ and that’s their sole purpose,” explained Emily Bender, a computational linguist at the University of Washington. “The resulting sequence of words may resemble expected discourse in that context but lacks reasoning, understanding, or factual analysis.”
Due to this characteristic, slight modifications in question phrasing can yield different results. Experts also noted that users could influence the model towards desired answers. Gilliard highlighted that these systems often display what developers refer to as “sycophancy,” where they tend to agree with or appease the user.
“Frequently, when users provide criticism or correction to the machine, it responds with phrases like ‘Oh, I’m sorry’ or ‘You’re right,’ making it possible to coax these systems into delivering preferred answers,” he added.
T. Philip Nichols, a Baylor University professor examining technology’s impact on education, remarked that keyword searches offer minimal insight into the actual teaching of a subject. He described the tool as “a blunt instrument” incapable of comprehending how certain discussions, which the software might flag as irrelevant, integrate into broader course themes.
“Pedagogical choices made by instructors might not be captured in a syllabus, so simply querying a chatbot to check if a topic is mentioned fails to reveal how it is discussed or presented,” Nichols stated.
Castillo’s account of testing the AI tool was the only instance in the reviewed records where Texas A&M administrators mentioned specific search terms employed to assess course content. In another communication, Castillo mentioned that she would share search terms with staff either in person or via phone rather than through email.
Officials from the system did not provide the list of search terms intended for the audit.
Martin Peterson, a philosophy professor at Texas A&M who focuses on technology ethics, pointed out that faculty have not been consulted regarding the tool, including members of the university’s AI council. He indicated that the council’s ethics and governance committee is responsible for establishing standards for responsible AI usage.
While Peterson generally opposes the initiative to audit the university system’s courses, he expressed some openness to the potential use of such tools.
“It’s essential that we thoroughly evaluate the tool before implementation,” Peterson remarked.
At Texas State University, officials directed faculty to revise their syllabi and suggested utilizing AI for this purpose.
In October, administrators identified 280 courses for review, instructing faculty to modify titles, descriptions, and learning outcomes to eliminate language deemed non-neutral. Records show that numerous courses slated for the Spring 2026 semester from the College of Liberal Arts were flagged for neutrality issues, including Intro to Diversity, Social Inequality, Freedom in America, Southwest in Film, and Chinese-English Translation.
Faculty were given a deadline of December 10 to complete their revisions, with a subsequent review scheduled for January and a full evaluation of the catalog expected by June.
Administrators provided faculty with a guide outlining language that they claimed indicated advocacy. It discouraged learning outcomes that require students to “measure or require belief, attitude, or activism (e.g., value diversity, embrace activism, commit to change).”
Additionally, administrators supplied a prompt for faculty to input into an AI writing assistant alongside their materials. This prompt instructs the chatbot to “identify any language that signals advocacy, prescriptive conclusions, affective outcomes, or ideological commitments” and to generate three alternative versions that omit those elements.
Jayme Blaschke, assistant director of media relations at Texas State, characterized the internal review as “thorough” and “deliberative,” but declined to disclose whether any classes have already been revised or removed, stating only that “measures are in place to guide students through any adjustments and maintain their academic progress.” He also refrained from explaining how courses were initially flagged and who established the neutrality criteria.
Faculty members have expressed that these changes are altering the decision-making process regarding curriculum on campus.
Aimee Villarreal, an assistant professor of anthropology and president of Texas State’s chapter of the American Association of University Professors, noted that curriculum decisions are typically faculty-led and evolve over an extended timeframe. She believes that the structure of this audit enables administrators to more closely supervise how faculty articulate their
