diff --git a/CHANGELOG.markdown b/CHANGELOG.markdown
index 0d576cc..866a4a2 100644
--- a/CHANGELOG.markdown
+++ b/CHANGELOG.markdown
@@ -1,3 +1,11 @@
+next [????.??.??]
+-----------------
+* `Numeric.AD.Mode.Reverse.Double` now handles IEEE floating-point special
+  values (e.g., `NaN` and `Inf`) correctly when `ad` is compiled with `+ffi`.
+  Note that this increase in floating-point accuracy may come at a slight
+  performance penalty in certain applications. If this negatively impacts your
+  application, please mention this at https://github.com/ekmett/ad/issues/106.
+
 4.5.4 [2023.02.19]
 ------------------
 * Add a `Num (Scalar (Scalar t))` constraint to `On`'s `Mode` instance, which is