Yea this makes sense to me; if they were just analyzing student commit history, but 'hiding & executing code' is conceptually dangerous, facilitates their surveillance, even if they might say "It's just git commit & git push".
its realistic to assume anthropic (and other frontier labs) have already gobbled the entire surface of the public internet into training data, plus so much more llm-generated traning data, that if you have public works and a model has enough parameters, i guess you are now 'in the models knowlege' the same way a figure like michael jackson or elon musk would be 'baked into the weights of the model' scary.
Sort of, I'm hesitant to say surveillance, since its less of 'running spyware' or something similar and more 'tracking student commit history'; where it gets weird is this section of the paper:
In our system, the Makefile or Project file that compiles the
project contains Git commit and push commands to automatically
commit changes into the student repository. Using this system,
changes are tracked every time the project is compiled. When a
student modifies a source file as a part of the program-build-test-
debug cycle, the Makefile commits and pushes the recent changes
into source control. This creates a fine-grained sequence of commits
that tell the story of how the program was developed.
They basically force-commit to your repo whenever you build your code, so they are able to 'track' your development?
If anyone is interested, the professor, Jeff Turkstra, wrote a paper called 'Tracking Large Class Projects in Real-Time Using Fine-Grained
Source Control' https://turkeyland.net/research/encourse.pdf and it's the suspected way the students got found out.