Reading the theorem is much simpler than reading the full proof - so the job here is to read the formalization of the theorem itself not the proof.
Then you depend on the Lean kernel for proof validation and on trusting it and that’s a different question.
Finally the question was also true for ‘non formalized math’ where you always depended on a trusted kernel of mathematicians who understood the problem well enough to sign off the claim