on the flip side: these models can create all the security holes they want in any given open source project they desire as long as they can convince a lazy AI llm to let them contribute code.

Everything you described can go through the black mirror and nullify it.

So, we're still here.