Personally, I'd much prefer proactive auditing (i.e. requiring source code) over waiting until someone is injured and possibly killed to launch an investigation.

Transparency is a dependency of trust. There's no going around that. If a product is not transparent, then it is not trustworthy, period. I'm not at all inclined to wait for something to violate my trust further before demanding a prerequisite for that trust. Governments shouldn't be so inclined, either, especially when it comes to human safety.

I'd also like that, but even today there is lots of software that can kill people if it contains bugs. It is usually enough that the makers can prove that they followed some process like EN50128. Source code is not required. I don't think there are competent people available for auditing source code.

I agree. I just don't think we should collectively allow companies to hide their source code and still be considered trustworthy.

— especially for innovative safety algorithms that could save the lives of their competitors' customers, if only the code was reusable.

Competent auditors will emerge if there's demand. There were once no competent astronauts!

"I don't think there are competent people available for auditing source code."

Probably not, but by requiring companies to disclose source code on request, effective independent audit of that code is merely improbable versus effectively impossible.