If researchers were unable to trust their collaborators, it would mean that they would have to master and oversee every step of every process. This would stop interdisciplinary research completely and massively reduce output.

In software, this would be the equivalent of not importing a package unless you first checked the code line by line.

The system works okay now, but there's a lot of room for improvement. I generally think the best way to improve it is to require more open data so that consumers of research can validate the findings of papers.