The OpenSSL Random Number generator documented here uses an OS-provided entropy source coupled with a MD5 hash function.
Does it impact the quality of randomness to use a broken cryptographic hash such as MD5 for random number generation? The topic is partially addressed in this question, with a focus on SHA1 rather than MD5 tough.
And if any impact, of which nature? Is it a risk of the RNG being predictable?
Back to OpenSSL: Though selecting another hash function seems to be doable at compilation time, there is no (apparent) configuration setting to change the MD5 to SHA2. So if there's any impact, I'd expect that most OpenSSL instances today are affected.