Deterrence layers failed because seeds changed parameters rather than task structure, scaffolding exposed solution steps, and hidden grading verified output consistency rather than authorship or understanding¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q1
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q1 i? Deterministic personalization allowed ChatGPT to produce parameterized solutions because seeds altered parameters rather than creating new reasoning problems. → Alexei Kaltchenko and Gurnivaj Tiwana 2026 q1 i? Grader acceptance established consistency with expected outputs, not independent authorship or understanding, because graders checked fields and reflection presence without semantic validation of notebook circuits. → Alexei Kaltchenko and Gurnivaj Tiwana 2026 (2)

Evidence¶

Alexei Kaltchenko and Gurnivaj Tiwana 2026¶

Alexei Kaltchenko and Gurnivaj Tiwana. (2026). ChatGPT Solves All Tested Qiskit Homework Assignments. arXiv. https://arxiv.org/abs/2608.19707 doi:10.48550/arxiv.2608.19707

q1 · i? · design · r2

The authors' interpretive analysis of the observed failure mode, not a tested comparison. For HW1, bit flips, the measurement map, and the expected count key were deterministic functions of student-visible data, so ChatGPT could construct both the exact output and the required circuit once it parsed those functions.

"A deterministic seed prevents every student from receiving exactly the same numeric configuration. It does not necessarily create a new reasoning problem."

Alexei Kaltchenko and Gurnivaj Tiwana 2026 (2)¶

Alexei Kaltchenko and Gurnivaj Tiwana. (2026). ChatGPT Solves All Tested Qiskit Homework Assignments. arXiv. https://arxiv.org/abs/2608.19707

q1 · i? · design · r2

The authors' analysis of the grading layers: hidden deterministic references made assignments autogradable but did not prevent ChatGPT from deriving correct answers from the public specification; optional QPY export was not a mandatory gate.

"Consequently, grader acceptance established consistency with expected outputs, not independent authorship or understanding."

Discussion¶

Related Claims¶

• ChatGPT produced executed, grader-accepted submissions for all three fixed personalized Qiskit assignment instances in 150 of 150 sessions — a narrower finding that bears on this claim
• Moral Unease, reported by 190 participants, reflects authorship guilt and a gap between AI output quality and actual understanding — related
• Reviewed literature reports risks of overreliance, AI errors in complex tasks, and students' uncritical acceptance of AI-generated outputs — related
• Indirect prompt injections combining instruction manipulation, role-playing, and obfuscation altered AI-assigned grades with attack success rates of 94–100% while remaining undetected — related
• Basic prompt injection attempts failed or were detected: instruction manipulation alone failed in all 9 iterations, role-playing without concealment was detected in output, and metadata-hidden injections failed in all 9 iterations — related