Purdue Instructor Addresses CS 240 AI Cheating Controversy

Overview

A Purdue University instructor has released a detailed retrospective addressing the academic integrity controversy that unfolded during the Spring 2026 semester of CS 240: Programming in C. The course, a cornerstone of the computer science curriculum, became the center of a significant debate regarding the use of generative AI tools by students to complete assignments.

Instructor Jeff Boles stated that his initial approach to handling the situation was flawed and acknowledged that the consequences for students found violating the policy were insufficient. He published the retrospective on September 25, 2026, aiming to correct what he described as widespread misinformation and inaccuracies circulating among the student body and the broader academic community.

"My approach to addressing the issue should have been better and is primarily why the individuals that ran afoul of the clearly stated course policy ultimately incurred little to no consequence."

The incident highlights the growing tensions between traditional academic integrity standards and the rapid integration of large language models (LLMs) into undergraduate education. Boles emphasized that the rules were explicitly communicated throughout the semester, contradicting claims by some students that they were unaware of the restrictions.

Policies and Expectations

Boles asserted that the prohibition against using AI to solve assignments was unambiguous. The course syllabus, available to all enrolled students, contained a specific clause stating that utilizing ChatGPT or similar software to programmatically generate solutions constituted academic dishonesty. Violations of this policy carried the penalty of an automatic failing grade for the course.

The syllabus also outlined a unique enforcement strategy. Instructors reserves the right to delay informing students of suspected violations, choosing instead to accumulate evidence across multiple assignments. This method allows for a more comprehensive understanding of the scope of cheating but requires careful documentation and discretion.

Furthermore, Boles noted that the expectation was reinforced repeatedly in lectures. Slides from Lecture 1 on January 12, 2026, explicitly placed the responsibility on students to read and understand the syllabus. Subsequent lectures on January 14, January 21, and February 4 included additional verbal and visual reminders regarding academic integrity.

Despite these repeated disclosures, some students implied that the policy was a surprise. Boles described such claims as specious, pointing out that the expectations were addressed in at least five separate lectures. He maintained that while some slides served as starting points for discussion, the core message regarding AI usage was consistent and clear.

The Argus Tool

A central component of the controversy was the use of a custom detection system named Argus. Misinformation suggested that the department had trained a proprietary large language model to identify cheaters. Boles clarified that this was not the case.

Argus utilizes static analysis rather than generative AI. The tool scans student source code for specific indicators that are highly unlikely to appear in legitimately written programs. These indicators serve as proxies for AI-generated code, such as unusual coding patterns or structural anomalies that differ significantly from human-written solutions.

The development of Argus was documented in a paper authored by the team over the summer. Although the paper was submitted to SIGCSE 2027 and was not selected for publication, the peer review process and results provided valuable insights. Preprints of the paper are available on arXiv, and the reviews can be accessed online.

Data analyzed using Argus on earlier semesters showed that the specific indicators used by the tool were virtually nonexistent prior to 2024. Their presence increased rapidly in subsequent years, aligning with the widespread adoption of LLMs in coding education. Boles noted that the tool demonstrated high accuracy in generating an H-score, a metric used to evaluate the likelihood of AI involvement.

"Providing such would almost certainly reduce the utility of the tool."

While the full list of indicators was not disclosed to protect the tool's effectiveness, Boles invited educators with legitimate interests to contact him directly. He emphasized that Argus served only as a starting point for investigation, not as a final verdict.

Investigation and Consequences

The deployment of Argus followed a conservative and thorough review process. Each potential case flagged by the tool was examined by at least one human reviewer before any action was taken. Boles stressed that not all flagged cases were pursued; only those with clear evidence and no reasonable explanation for the detected anomalies were escalated.

The AI Academic Integrity (AIAI) team was formed on March 23, 2026, to manage the high volume of potential violations. The team consisted of teaching assistants and instructors who worked to standardize the review process. Historical data from previous semesters, such as CS 159, provided a framework for how to handle these cases delicately and effectively.

Out of 584 students initially identified by the system, only 267 were pursued for formal review. This represents approximately 45.7% of the initial flags. The remaining cases were dismissed after human reviewers determined that the evidence was insufficient or that there were plausible non-cheating explanations for the code similarities.

Despite the rigorous review, Boles admitted that the outcomes were less severe than they should have been. The decision to prioritize education and rehabilitation over strict punishment, combined with procedural delays, resulted in minimal consequences for many students. He expressed regret that this approach may have undermined the deterrent effect of the academic integrity policy.

The retrospective serves as both a record of the events and a cautionary tale for other institutions facing similar challenges. As AI tools become more sophisticated, educators must balance the need for fair detection with the ethical responsibilities of academic oversight. Boles concluded that while the technology for detection has advanced, the human element of policy enforcement requires greater attention and consistency.