A degree must certify what a student can do without AI

Cheating controls answer only one question

Universities are reviving supervised exams, blue books, oral assessments, and device limits because unrestricted generation makes unsupervised authorship difficult to verify. Those methods can establish that a student completed a particular task without hidden assistance.

They do not define the capability represented by the full degree. A graduate may need to work independently in one situation and use AI effectively in another. Institutions need a curriculum and transcript that make the difference explicit rather than forcing one grade to stand for both.

A published rule creates a clear default

The University of Chicago Law School's published policy prohibits generative AI during examinations and requires every word of an exam answer to be produced by the student. It allows some study uses and lets instructors adopt different course-specific rules, while warning against entering confidential or personal information into AI systems.

That is a concrete standard, not a universal solution. It defines authorship for assessments, but professional competence also includes source verification, judgment, confidentiality, error detection, and the ability to explain an AI-assisted conclusion under pressure.

The downstream user bears credential uncertainty

Employers, clients, courts, patients, and public agencies rely on degrees as signals that a graduate can perform. If a program measures only polished artifacts without verifying independent reasoning, the institution transfers uncertainty to everyone who later trusts the credential.

That risk will not be distributed evenly. Prestigious institutions may preserve intensive human instruction and oral evaluation, while resource-constrained programs replace teaching with automated assistance. A common degree label could then conceal radically different levels of independently demonstrated skill.

Certify two capabilities honestly

Programs should create protected independent-performance assessments and separate AI-augmented assessments that test tool selection, prompting, source validation, confidentiality, error correction, and human responsibility. Both belong in modern education, but they should not be treated as interchangeable.

The strongest response to AI cheating is not a surveillance arms race. It is a credential system precise enough to say what the graduate can do alone, what the graduate can do with tools, and which claims the institution has actually verified.

• Use oral, supervised, and transfer tasks to verify independent reasoning.
• Assess AI use through source checks, error correction, judgment, and disclosure.
• Report the assessment mode for consequential capabilities.
• Preserve human teaching and feedback as core educational infrastructure.