I wonder if assignments are fixable. It might be that they become AI benchmarks which results in models optimized to solve them...

There is no fixing students who just want results.

In the end in-person examinations is probably the solved solution we have. Either verbal, on paper, or via very locked down computers.