Matter of the

This opinion is nonprecedential except as provided by
Minn. R. Civ. App. P. 136.01, subd. 1(c).

STATE OF MINNESOTA
IN COURT OF APPEALS
A25-0342

Matter of the

University of Minnesota,
Respondent,

vs.

Haishan Yang,
Relator.

Filed February 2, 2026
Affirmed
Reyes, Judge

University of Minnesota
File No. 2024029701

Douglas R. Peterson, General Counsel, Dan Herber, Associate General Counsel,
University of Minnesota, Minneapolis, Minnesota (for respondent)

Haishan Yang, St. Paul, Minnesota (self-represented relator)

Considered and decided by Reyes, Presiding Judge; Larkin, Judge; and Kirk,
Judge.∗
NONPRECEDENTIAL OPINION
REYES, Judge

∗ Retired judge of the Minnesota Court of Appeals, serving by appointment pursuant to
Minn. Const. art. VI, § 10.
2
In this appeal following a university disciplinary process, relator argues that
respondent failed to follow its internal procedures, violated his substantive- and
procedural-due-process rights, and made findings that were not supported by substantial
evidence. We affirm.
FACTS
Relator Haishan Yang was a Ph.D. student at respondent University of Minnesota.
In August 2024, relator took an exam with instructions permitting the use of class materials
but prohibiting the use of “any sort of Artificial intelligence tools, such as ChatGPT.”
A four-person faculty grading committee (the graders) reviewed relator’s exam
answers, and all four were concerned that relator had used some sort of artificial
intelligence (AI) software on the exam. One grader ran the exam questions through an AI
program and found the program’s answers (the AI-generated evidence) comparable to
relator’s answers. Another grader discussed the accusations with relator and then filed a
report with the Office for Community Standards (the office). The report alleged that relator
used AI to produce his answers, violating scholastic dishonesty provisions of the Student
Code of Conduct (the code), and it recommended “dismissal from the program as a
sanction.”
The office sent a letter to relator explaining the allegations. After meeting with the
office director, relator sent a 35-page response denying the allegations. The office sent an
updated letter (the resolution letter) suggesting an informal resolution of expulsion with
the option for a formal hearing at relator’s request. Relator chose the formal hearing and
requested representation by an advocate from respondent’s Student Advocate Services.
3
The Campus Committee on Student Behavior (the committee) held a formal hearing
via Zoom on November 15, 2024. The relevant individuals at the hearing included: the
chair of the committee, serving as the hearing chair; six voting panel members, serving as
the hearing panel; the office director , representing respondent; the assistant director of
respondent’s Student Advocate Services, representing relator; and two graders as well as
relator’s academic advisor, attending as witnesses.
During the hearing, b oth parties gave opening statements, presented exhibits ,
examined and cross-examined witnesses, and gave closing statements. Respondent’s
exhibits included the AI-generated evidence, which it compared to relator’s exam answers.
In the middle of the hearing, relator’s representative asked one of the graders to generate a
new exam answer from ChatGPT. Respondent’s representative intervened, asking for a
breakout room with both representatives and the hearing chair. The hearing transcript
indicates that these individuals held a conversation off the record. When they returned to
the main Zoom room, the hearing chair explained:
We just had a brief discussion there, and we ’re just going to
cease this particular line of questioning and disregard any
requests that you guys do your own generation of responses
through ChatGPT. I just want the panelists to base their
deliberations on the evidence that’s presented today. We don’t
create new evidence.

Following the hearing, the committee issued a disposition letter setting out the
hearing panel’s findings and recommendations.
1 The letter explained that the hearing panel

1 The disposition letter notes that the hearing panel’s findings and recommendations are
based on the deliberations of five panel members because one had to leave due to an
emergency.
4
unanimously found that relator violated the code and had “more likely than not” used AI
“to develop answers for the exam.” The hearing panel did not mention the AI-generated
evidence as support for its decision. Instead, it: “credited the ability of the testifying
[graders] to identify AI generated papers ”; found that “irrelevant” sources referenced in
relator’s exam “raised significant questions about the use of AI”; and found problems with
relator’s “lack of citations, a pattern of continued excuses about exam errors, and
inconsistencies in [his] testimony.” Relying on one grader’s testimony that the trust
required between doctoral students and their professors “has been destroyed” because of
relator’s AI use, the “panelists agreed that it would be extremely difficult for that trust to
be reestablished.” The disposition letter directed relator’s expulsion.
Relator appealed this decision to the provost’s office, and an appellate officer there
affirmed the hearing panel’s findings and sanction.
This appeal follows by writ of certiorari.
DECISION
Relator appeals by writ of certiorari under Minnesota Statutes sections 606.01 to
606.06 (2024). See also Brenny v. Bd. of Regents, 813 N.W.2d 417, 421 (Minn. App. 2012)
(requiring petition for writ of certiorari to review discretionary university decision).
Judicial review of a state agency, like respondent, is “both limited and deferential.”
Williams v. Smith, 820 N.W.2d 807, 813 (Minn. 2012). “Appellate courts generally defer
to university decisions in student discipline matters.” Tatro v. Univ. of Minn., 800 N.W.2d
811, 815 (Minn. App. 2011), aff’d on other grounds, 816 N.W.2d 509 (Minn. 2012).
However, appellate courts will sustain a certiorari challenge if: (1) a university decision “is
5
based on an error of law, or . . . is arbitrary, oppressive, unreasonable, or without evidence
to support it,” or (2) the university failed to “engage in reasoned decision-making” or
“affirmatively state legally sufficient reasons for its decisions that are factually supported
in the record.” Williams, 820 N.W.2d at 813. For issues that require constitutional
interpretation, we apply de novo review. Star Trib. Co. v. Bd. of Regents, 683 N.W.2d 274,
283 (Minn. 2004).
Relator raises 15 issues, which we consolidate into five categories: (1) issues that
are not properly before this court; (2) relator’s substantial-evidence claim; (3) his claim
that respondent failed to follow its own procedures; (4) relator’s procedural-due-process
claim; and (5) his substantive-due-process claim.
I. Relator failed to preserve four issues.
“Generally, we will not consider an issue not addressed below.” In re A.D., 883
N.W.2d 251, 261 (Minn. 2016) (quoting Thiele v. Stich, 425 N.W.2d 580, 582 (Minn.
1988)). Rather, this court’s review is “limited to an inspection of the record of the
administrative tribunal.” Chronopoulos v. Univ. of Minn., 520 N.W.2d 437, 441 (Minn.
App. 1994), rev. denied (Minn. Oct. 27, 1994). But appellate courts have discretion to
address issues not raised below. See A.D., 883 N.W.2d at 261.
Here, relator argues for the first time on appeal that: (1) the office director displayed
“partiality”; (2) the reporting grader violated respondent’s reporting procedures; (3) the
hearing chair improperly allowed two of respondent’s witnesses to attend the hearing rather
than restrict respondent’s witnesses to one; and (4) respondent prevented relator from
“confront[ing] his accusers” because only two of the four graders attended the hearing.
6
Relator did not present these issues to the hearing panel or the provost’s office. We
therefore decline to consider these four issues.
II. Respondent’s findings were supported by substantial evidence.
Relator argues that respondent’ s findings were not supported by substantial
evidence because the hearing panel’s decision did not cite to “specific evidence” and only
generally refers to relator’s exam answers and testimony. We are not persuaded.
Appellate courts review quasi-judicial decisions under the substantial-evidence test.
See In re Brown, 28 N.W.3d 486, 505 (Minn. App. 2025), rev. denied (Minn. Jan. 21,
2026). “Substantial evidence is 1) such relevant evidence as a reasonable mind might
accept as adequate to support a conclusion; 2) more than a scintilla of evidence; 3) more
than some evidence; 4) more than any evidence; and 5) evidence considered in its entirety.”
Id. (quotation omitted).
The hearing panel explained that it found respondent’s presentation “compelling”;
“credited the ability of the testifying [graders] to identify AI generated papers”; “agreed
that the documents used in the exam that were demonstrated to be irrelevant to the exam
questions raised significant questions about the use of AI”; and found problematic “the
lack of citations” in relator’s exam, relator’s “pattern of continued excuses about exam
errors,” and “inconsistencies in [relator’s] testimony.” After a careful review of these
findings and the underlying record, we conclude that substantial evidence supports the
hearing panel’s determinations that relator violated the code and more likely than not used
AI to develop his exam answers.
III. Respondent followed its internal procedures.
7
Relator argues that respondent failed to follow its internal procedures—the
Committee Hearing Procedures (the hearing procedures), the code, and the Student
Conduct Code Procedure (the code procedure) 2—because (A) the hearing chair held an
off-record conversation, making judicial review impossible; (B) the hearing chair admitted
the AI-generated evidence; (C) the office failed to include findings in its resolution letter;
(D) the appellate officer reject ed alleged “new evidence”; and (E) the appellate officer
substituted his own judgment for that of the hearing panel. We disagree.
A. The record is adequate to permit review, and the hearing chair did not
violate the hearing procedures by holding an off-record conversation.

Relator argues that an off-record discussion made judicial review impossible and
violated the hearing procedures.
Judicial review is available even when the record “plainly falls short of a record
generated in formal judicial proceedings.” Dietz v. Dodge County, 487 N.W.2d 237, 240
(Minn. 1992). If the record is “adequate to ascertain” the answer to the relevant issue(s)
on appeal, we can review the administrative decision. Id.; cf. Honn v. City of Coon Rapids,
313 N.W.2d 409, 415-16 (Minn. 1981) (holding nonverbatim record of public proceedings
was sufficient when reasons for decisions were put in writing and not made in “conclusory
fashion”). Although the exact words exchanged in the breakout room here are not available
for us to analyze, the entire administrative record provides adequate information to permit
our review. The transcript shows that relator asked to generate a new exam answer using

2 Student Conduct Code Procedure: Twin Cities, Univ. of Minn.: Off. of Institutional
Compliance, https://policy.umn.edu/education/studentconductcode -proc01
[https://perma.cc/TZ8E-JRUT].
8
ChatGPT. Respondent asked for an off-the-record discussion with the representatives and
the hearing chair. The hearing chair then stated on the record that it would disregard
relator’s request and have the panel focus on the evidence in the record without creating
new evidence. We conclude that the record is more than adequate to ascertain the relevant
issues, permitting our review of the hearing chair’s determination.
We are also not persuaded that permitting the off-record discussion violated the
hearing procedures. The hearing procedures only prohibit off-record conversations
between panelists and the parties or representatives, not between the representatives and
the hearing chair. Relator’s reliance on Ganguli v. University of Minnesota, 512 N.W.2d
918 (Minn. App. 1994), is misguided. In Ganguli, a university committee “violated its
own procedures” by failing to make any findings when terminating a professor’s
probationary appointment. 512 N.W.2d at 920, 923. Here, the off-record conversation did
not violate any university procedures.
B. The hearing chair did not violate the hearing procedures by making
certain evidentiary determinations.

Relator next argues that respondent violated the hearing procedures by interrupting
him, limiting his cross -examination, imposing time limits, and allowing the admission,
over his objections, of “unreliable and prejudicial evidence,” namely the AI-generated
evidence, “a dismissed allegation” of previous scholastic dishonesty, and an AI-probability
score.
The hearing procedures task the hearing chair with “maintaining an orderly, fair,
and respectful hearing,” “respond[ing] to disruptive or harassing behaviors, ” and
9
“impos[ing] reasonable time limits on any phase of the proceedings.” The hearing
procedures also permit each party to “ offer reliable information relevant to the issue ,”
“object to the information offered by the other party,” and “introduce relevant written
documents, objects, films, or other materials as exhibits.” Ultimately, however, the hearing
chair and the hearing panel “have discretion t o determine what information should fairly
be included or excluded” in the hearing.
While relator made certain evidentiary objections, the hearing chair disagreed with
relator’s characterization of respondent’s evidence and admitted the challenged evidence.
Relator then had the opportunity to present his own evidence and testimony, at which point
he explained that a previous allegation of scholastic dishonesty was dropped, AI-generated
output is unreliable evidence both as a general rule and in this specific case, and AI-
probability scores are not trustworthy or consistent. The hearing panel considered the
presentations of both parties. We conclude that the hearing chair did not violate the hearing
procedures.
C. Respondent’s resolution letter did not violate the code.

Relator argues that respondent violated the code by failing to include “factual
findings” in its resolution letter.
The code sets out the hearing process for charges of student-conduct violations. It
requires, in part, that “[a] finding of responsibility for violation of the [code] must be based
on a preponderance of the evidence.” After making this finding of responsibility,
respondent may offer the student a “proposal of an informal resolution that includes the
findings and outcomes being offered to resolve the incident.”
10
In the resolution letter here, respondent explained the basis of the accusations,
including that each grader reviewed relator’s exam “independently,” “shared that
[relator’s] answers seemed to not be directly relevant to the question, or involved concepts
that were not covered in class or the readings,” “had significant concerns that some sort of
[AI] was used in the answers,” and noted that an acronym used by relator was “not
commonly used in the field” or by the graders. We conclude that respondent provided
findings in its resolution letter and thus did not violate the code.
D. The a ppellate officer did not violate the code procedure by rejecting
relator’s claims of “new evidence.”

Relator claims that the appellate officer improperly rejected “new evidence” that the
graders altered the AI-generated evidence before sharing it with colleagues.
Under the code procedure, a student can appeal a disciplinary decision if “[n]ew
evidence exists [(1)] that was not previously available to the appealing party and [(2)] that
is sufficient to affect the outcome.”
Here, the appellate officer stated that “[t]he record is not clear” as to the first prong
but, regardless, relator’s appeal failed on the second prong. The appellate officer explained
that relator’s alteration allegations were not credible and that, even if true, the alleged
alterations “are such that they would not have impacted the analysis.”
As discussed above, the hearing panel did not rely on the AI-generated evidence to
support its findings. We conclude that the appellate officer did not err by determining that,
even if the evidence was “new” and the alteration allegations were true, that would not be
“sufficient to affect the outcome” of the hearing.
11

12
E. The appellate officer did not violate the code procedure by determining
that the hearing panel relied on substantial information.

Relator argues that the appellate officer violated the code procedure by “substituting
his own evidence and factual findings” for that of the hearing panel.
The code procedure instructs that an appellate officer reviewing a substantial-
information claim must “respect the credibility determinations of the hearing body and
must not substitute the officer’s judgment for the hearing body.” The appellate officer’s
task is merely to “determine whether the hearing body’s decision was unreasonable (i.e.,
arbitrary) in light of the information presented.”
The appellate officer here recognized and outlined the information presented to the
hearing panel. The appellate officer carefully reviewed that information and, without
substituting his own judgment, concluded that substantial information supported the
hearing panel’s findings.
In sum, we conclude that respondent did not violate any of its internal procedures.
IV. Respondent did not violate relator’s procedural-due-process rights.
Relator next argues that respondent violated his procedural-due- process rights
because it withheld until the end of the hearing a link to the AI-generated evidence, which
had content that differed slightly from the AI -generated evidence presented earlier in the
case. According to relator, this “depriv[ed] him of adequate notice of the charges and a
meaningful opportunity to respond.” This argument is unavailing.
“Whether the government has violated a person’s procedural -due-process rights is
a question of law that this court reviews de novo.” ITW Food Equip. Grp. LLC v. Minn.
13
Plumbing Bd. , 933 N.W.2d 523, 531-32 (Minn. App. 2019). When evaluating a
procedural-due-process claim, appellate courts first “identify whether the government has
deprived the individual of a protected life, liberty, or property interest,” and then, if that
type of deprivation exists, we “determine whether the procedures followed by the
government were constitutionally sufficient.” Sawh v. City of Lino Lakes, 823 N.W.2d
627, 632 (Minn. 2012) (quotation omitted).
Expulsion from a university meets the first element of the procedural-due-process
analysis. See Abbariao v. Hamline Univ. Sch. of L., 258 N.W.2d 108, 112 (Minn. 1977)
(“[A] student’s interest in attending a university is a property right protected by the due
process clause.”); see also Shuman v. Univ. of Minn. L. Sch., 451 N.W.2d 71, 74 (Minn.
App. 1990) (reiterating this principle), rev. denied (Minn. Mar. 16, 1990). We therefore
turn to an analysis of the second element: whether the procedural protections afforded to
relator were constitutionally sufficient.
To analyze procedural-due-process claims, Minnesota appellate courts “ have
consistently applied” the three-part test outlined in Mathews v. Eldridge, 424 U.S. 319, 335
(1976). Olson v. One 1999 Lexus, 924 N.W.2d 594, 603 (Minn. 2019). Courts consider:
First, the private interest that will be affected by the official
action; second, the risk of an erroneous deprivation of such
interest through the procedures used, and the probable value, if
any, of additional or substitute procedural safeguards; and
finally, the Government’s interest, including the function
involved and the fiscal and administrative burdens that the
additional or substitute procedural requirement would entail.

Mathews, 424 U.S. at 335.
14
Under the first Mathews factor, the private interest affected by respondent’s decision
is relator’s interest in attending a university. This factor therefore weighs in favor of
relator.
The second Mathews factor considers “ the risk of an erroneous deprivation of a
protected interest” and the likely benefit of “additional safeguards.” Sawh, 823 N.W.2d at
634. Relator asserts that he was not provided with meaningful notice or a reasonable
opportunity to be heard. We are mindful that, as relator notes, respondent did not provide
a direct link to the AI-generated evidence until the end of the hearing. But relator had
notice of the existence of the AI-generated evidence by at least September 9, 2024, because
he disputed its validity in a letter on that date. Respondent also provided notice of the
charges against relator as early as September 3, 2024, when respondent sent its first letter
to relator explaining the allegations against him, the basis of the graders’ concerns, and the
portion of the code allegedly violated.
“Minnesota courts have consistently recognized that the basic rights of procedural
due process required in quasi-judicial proceedings such as this are satisfied by ‘reasonable
notice of hearing and a reasonable opportunity to be heard.’” Brown, 28 N.W.3d at 501
(quotation omitted)); see also Shuman, 451 N.W.2d at 74 (requiring “notice and some
opportunity to be heard” in student disciplinary process).
The record shows that respondent afforded relator both “reasonable notice of
hearing and a reasonable opportunity to be heard.” Brown, 28 N.W.3d at 501 (quotation
omitted). In addition to notice of the charges and the evidence discussed above, relator
received an evidentiary hearing at which he called a witness, cross-examined witnesses,
15
presented evidence, and disputed the validity of respondent’ s evidence. This factor
therefore weighs in favor of respondent. Cf. Shuman, 258 N.W.2d at 75 (concluding
students received sufficient procedural due process through “a full evidentiary hearing with
a right to counsel and the chance to call witnesses and cross-examine witnesses”).
Even further, there is no evidence in the record “indicating any ‘probable value . . .
of additional or substitute procedural safeguards.’” Brown, 28 N.W.3d at 502 (quoting
Mathews, 424 U.S. at 335). Indeed, while relator claims that the delayed access to the link
“had a material impact on the outcome of the hearing,” the record shows that relator was
on notice of the existence of the AI-generated evidence months before the hearing, he had
access to a copy of it before the hearing, he saw respondent present it during the hearing,
and his representative disputed its importance at the hearing. The hearing panel also did
not cite to the AI -generated evidence in its hearing decision and, as discussed below, any
alleged alterations in the linked version of the AI-generated evidence would not have
affected the hearing panel’s decision. Relator does not articulate how his access to the link
would have changed the hearing panel’ s decision and findings, and the record does not
support that contention . The lack of probable value of additional procedural safeguards
mitigates any risk of erroneous deprivation of relator’s interest. See id. at 502.
The third Mathews factor considers respondent’s interests, “including the fiscal and
administrative burdens that would be required to impose additional or substitute procedural
requirements.” Sawh, 823 N.W.2d at 635. Relator suggests that respondent must disclose
all potential evidence before a disciplinary hearing, or even before a resolution letter. But
this would impose significant administrative burdens on respondent and defy the general
16
principle that “quasi-judicial proceedings do not invoke the full panoply of procedures
required in regular judicial proceedings, civil or criminal, many of which would be plainly
inappropriate in these quasi-judicial settings.” Brown, 28 N.W.3d at 501 (quotation
omitted). This factor therefore weighs in favor of respondent.
Because two of the three Mathews factors weigh in favor of respondent, we
conclude that the process respondent afforded relator in making its quasi-judicial
determination was constitutionally sufficient and did not violate his procedural-due-
process rights.
V. Respondent did not violate relator’s substantive-due-process rights.
Relator argues that respondent violated his substantive-due-process rights because
the hearing panel considered “mistrust” in its recommendation of expulsion, the graders
altered the AI-generated evidence and exhibited bias against him, and the hearing chair
exhibited bias against him. We disagree.
To establish a violation of his substantive -due-process rights, relator must show
“that the University acted arbitrarily in dismissing him or that the dismissal was such a
substantial departure from accepted academic norms as to demonstrate that the person or
committee responsible did not actually exercise professional judgment.” Ross v. Univ. of
Minn., 439 N.W.2d 28, 34 (Minn. App. 1989) (quotation omitted), rev. denied (Minn. July
12, 1989). We address each element in turn.
A. Relator failed to demonstrate that respondent acted arbitrarily.

To support a substantive-due-process claim, “the government’s conduct [must be]
‘so egregious that it shocks the conscience.’” Brown, 28 N.W.3d at 503 (quoting Mumm
17
v. Mornson, 708 N.W.2d 475, 487 (Minn. 2006)). “Only the most extreme instances of
governmental misconduct satisfy this exacting standard.” Mumm, 708 N.W.2d at 487. The
code states: “Factors to consider in determining appropriate outcomes include: the nature
of the offense, the severity of the offense, the culpability of the student or student group,
the impact on other students or members of the University community, and the opportunity
for student development.” The code explains that, while expulsion “is a serious outcome,”
it may be appropriate for certain code violations, including “serious scholastic dishonesty.”
The hearing panel unanimously found that relator violated the scholastic dishonesty
provision of the code. It upheld the recommended outcome of expulsion, explaining:
“Panelists resonated with faculty statements that trust would be essential to continuing in
the program. However, after using AI on the exam, panelists agreed that it would be
extremely difficult for that trust to be reestablished. ” This explanation include d
consideration of at least two of the relevant factors above, including “the nature of the
offense” and “the impact on other . . . members of the University community.” Relator’s
disagreement with his expulsion does not make it conscience-shocking conduct. Relator
identifies no evidence that supports his claim that his expulsion was decided arbitrarily,
and the record does not support that contention.
B. Relator failed to establish that the hearing panel did not exercise
professional judgment.

Relator is “entitled to a decision by an unbiased decisionmaker.” Buchwald v. Univ.
of Minn., 573 N.W.2d 723, 727 (Minn. App. 1998), rev. denied (Minn. Apr. 14, 1998). But
18
“[t]here is a presumption of administrative regularity, and the party claiming otherwise has
the burden of proving a decision was reached improperly.” Id.
Relator does not identify any evidence that the hearing panel, as the decision-maker,
failed to exercise professional judgment. Instead, he focuses almost exclusively on
allegations of bias against the graders and the hearing chair, in addition to claims that the
graders altered the AI-generated evidence. Relator does not explain how any alterations to
the AI-generated evidence affected the hearing panel’s decision, which did not cite to or
credit that evidence.
In addition, relator does not explain how any bias of the graders or hearing chair
affected the hearing panel, which did not include those individuals. Even if the graders
were biased against him, leading to the initial charge of scholastic dishonesty, relator does
not explain how that bias might have influenced the hearing panel and its unanimous
decision. To support his claims of bias by the hearing chair, relator cites Gibson v.
Berryhill, 411 U.S. 564 (1973), claiming it stands for the proposition that he is entitled to
“an unbiased decision-maker free from conflicts.” But Gibson focuses on a federal district
court’s determination that a state’s board of optometry “was so biased by prejudgment and
pecuniary interest that it could not constitutionally conduct hearings” on whether to revoke
optometry licenses. Id. at 578. There, the board had a “personal interest” in revoking the
licenses of certain optometrists because of the board’s legal and financial endeavors. Id.
at 571, 578-79. The United States Supreme Court concluded: “[T]hose with substantial
pecuniary interest in legal proceedings should not adjudicate these disputes.” Id. at 579.
19
The facts here are very different. As discussed, the hearing chair did not have any
decision-making role in relator’s case and, as with the graders, relator presents no evidence
that any bias of the hearing chair affected the hearing panel and its decision. Relator also
presents no evidence that the hearing chair had any pecuniary interest in his disciplinary
proceeding, let alone a “substantial” one. Relator has not met his burden to overcome the
presumption of administrative regularity in seeking to establish a violation of his
substantive-due-process rights. We conclude that respondent did not violate relator’s
substantive-due-process rights.
Affirmed.