Please bear in mind that this information is all secondhand. I have not looked closely at the original drafts of Hash DRBG (although you might find a draft that's early enough if you peruse the FOIA results in ).
However, during conversations with folks at NIST I was told that there were certain weaknesses in early drafts of Hash DRBG that were very similar to known issues in the FIPS 186 (DSA) DRBG. Kelsey, Schneier, Wagner and Hall have at least one paper on this . In particular, if an attacker is able to control some of the entropy inputs to the DRBG, they could force that DRBG to repeat output. I gather from these discussions that the original draft of Hash DRBG had similar issues. The version published in NIST SP 800-90A was modified to deal with these.
It's important to note that I'm not saying these flaws were deliberate. Quite frankly they just look like bugs. And they appear to have been fixed long before to the publication of NIST SP 800-90A. I also can't state with certainty that Hash DRBG was an NSA design, though I believe the NSA was involved in its design.
One more note: it's my understanding that Hash DRBG was eliminated in the ANSI standard because it was deemed redundant, but it was kept in NIST SP 800-90 (though in repaired form) because the NSA asserted that some partners had already implemented it. Hash DRBG was also included in ISO 18031, some portions of which I can probably dig up and add to this link later. It's not clear whether the ISO version includes the repairs.
Welcome to our crazy world!