diff --git a/mock/py/mockbuild/backend.py b/mock/py/mockbuild/backend.py
index f9ad24d91..15a40f923 100644
--- a/mock/py/mockbuild/backend.py
+++ b/mock/py/mockbuild/backend.py
@@ -783,9 +783,6 @@ def get_command(mode):
                     # We want to (re-)write src.rpm with dynamic BuildRequires,
                     # but with short-circuit it doesn't matter
                     mode = ['-ba']
-                # rpmbuild -br already does %prep, so we don't need waste time
-                # on re-doing it
-                mode += ['--noprep']
 
         self.buildroot.doChroot(get_command(mode),
                                 shell=False, logger=self.buildroot.build_log, timeout=timeout,
diff --git a/releng/release-notes-next/generate_buildrequires.bug b/releng/release-notes-next/generate_buildrequires.bug
new file mode 100644
index 000000000..68d0204dd
--- /dev/null
+++ b/releng/release-notes-next/generate_buildrequires.bug
@@ -0,0 +1,17 @@
+To save time, Mock previously skipped the `%prep` section when building
+an RPM package after completing the installation of all the requirements
+generated by `%generate_buildrequries`.
+
+However, when the `%generate_buildrequries` section has side effects altering
+the results of a subsequent run of self,
+then the final build may work with different data and generate incorrect
+build requirements that may not be actually installed during the build.
+In extreme case, it is possible to successfully build a package with
+unsatisfiable BuildRequires in the built SRPM metadata.
+
+Mock now calls `%prep` section even during the final build after completing
+the installation of all the generated build requirements.
+This may result in a few seconds of delay when spec file uses
+`%generate_buildrequries` but the source used to generate the build requirements
+is more likely to stay consistent during the build,
+making it harder to accidentally generate unsatisfied BuildRequires.