I'm not saying it's impossible, but it's very hard in practice, and even harder to determine if things actually work, ofc it also depends on your threat model and adversary an old paper on the topic worth reading:
