This sort of thing should be trivial but unfortunately it is not. The best place to get this sort of information is in the host field of GenBank formatted sequences. It is missing for a large number of sequences but if you parse the information for the whole of Genbank you should be able to get a good representation of the known host-pathogen associations.

You might also find this previousl post useful: How To Determine If A Given Species Or Taxon Id Corresponds To A Pathogen Or Not? and Rod Page's blog

You will have to manually curate this information for species and isolates involved. You'll have to deal with zoonotic pathogens, as well as undocumented strains or isolates. Zoonotic pathogens prove to be troublesome because of the sometimes diverse set of hosts. Lastly, you need to catch things where an isolate was only reported in an animal, but the species in general is zoonotic, making it usually a human pathogen as well. This means text mining is especially tricky and could generate a ton of false negatives.

You might be able to find some pathogen databases, but this isn't something that will show up in NCBI and can't be done with sequence data. If you solved the problem of predicting pathogen genetic changes associated with species jumping, you'd revolutionize emerging pathogens research forever.

You might want to check the CDC's Biosafety in Microbiological and Biomedical Laboratories, it will cover a very large array of bacteria. However, the animal only pathogens will be limited to serious agricultural pathogens. So it have a far from complete list.

ABSA has a database you can check, but again this is geared towards infectious disease research and biocontainment, not bacteriology globally. http://www.absa.org/riskgroups/bacteria.html