HN user
This is a scary trend. I have seen quite a few repos with fully autogenerated code. Some sort of reporting would be required. On GitHub side one check that could expose such bad actors would be to monitor how much code is committed per time period (this particular guy did most of the code 2-3 months ago).
It is still early days - gpt powered browsers are becoming a commodity. One use I've found is leadgen automation. But it still not stable and frequently ends up in infinite loops.
I've started working on this new idea of building a developer-first, system-level AI compliance solution (within the context of EU AI Act) which would allow exporting regulator-ready evidence packages (AI audit trails - model versions, dataset quality metrics, performance evaluations, decision/human intervention logs etc. ) with one click.
Think Sentry for AI systems with the functionality of regulator friendly export bundles.
Tell me why this is a bad idea.