68% of Computer Science Instructors Have Changed Their Exams Because of AI, ACM Survey in 49 Countries Finds

In July 2026, the Association for Computing Machinery (ACM) published a survey of 763 computer science instructors in 49 countries. It found that 68% have already changed their exam format because of generative AI, moving toward oral exams, proctoring and project work.

What the ACM Survey Found

The ACM survey documents a sweeping rethink of how knowledge is assessed: more than two-thirds of computer science instructors have restructured their exams to account for the ubiquity of AI assistants. The shift is fundamental. The emphasis is moving from the ability to write code to the ability to read, understand and explain it.

• 763 computer science instructors from 49 countries
• 68% have already changed their exam format because of AI
• New formats: oral exams, proctoring, project work
• Almost half of respondents have no proven examples of integrating AI into courses
• The focus of teaching is shifting from writing code to understanding it

Why Instructors Are Changing Exams

Instructors are changing exams because AI now solves classic written coding assignments in seconds. Homework and standard tests no longer reflect a student's actual level: tools like ChatGPT, Claude and Copilot write working functions from a short prompt, and copying a solution has become easier than writing one.

In response, 68% of respondents have moved to formats that are harder to delegate to a model. An oral exam shows whether a student understands their own solution. Proctoring restricts access to AI during the work. A project defense tests whether a student can take a system through to a result and explain every decision along the way. In effect, universities are returning to live dialogue with students in places where automated code checking used to be enough.

What Is the Paradox of the AI Tutor?

The paradox is that the same AI that devalues old exams may be the strongest personal programming tutor ever built. It explains errors, walks through algorithms and offers hints step by step, but it also lets a student pass a course without ever learning to write code independently.

That is why instructors are scrambling to rethink how to assess real skills. And here the main problem surfaces: almost half of the ACM survey participants admit they have no proven examples of how to properly build AI into the teaching process. Each university and each course is, in effect, experimenting on its own, without a shared methodological foundation.

Teaching is shifting from writing code to understanding it. — finding

of the ACM survey of 763 instructors in 49 countries

What This Means

Higher education in computer science is reorganizing faster than teaching guidelines can be written: two-thirds of instructors have already changed their exams, but there is still no generally accepted recipe for teaching programming in the AI era. Universities will spend the next few years searching for formats in which AI helps students learn rather than replacing the learning itself, and the ACM survey is the first large-scale measurement of how fast this is happening.

Frequently Asked Questions

How many instructors took part in the ACM survey?

The ACM survey included 763 computer science instructors from 49 countries. It is one of the largest international snapshots of opinion on AI's impact on teaching programming.

How are universities changing exams because of AI?

According to ACM, 68% of instructors have moved to oral exams, proctoring and project work. These formats are harder to pass with the help of AI and do a better job of testing understanding of code rather than the ability to generate it.

What is preventing AI from being built into courses?

Almost half of instructors said they have no proven examples of integrating AI into teaching. Ready-made methods and standards for using models without damaging students' skills have not yet taken shape.