diff --git a/CHANGELOG.md b/CHANGELOG.md
index 68856034..384b40ab 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,7 @@
 ### Fixed
 - Don't print message about run being restored if the run is canceled or failed: this usually means that the module instance couldn't be started at all because the account tier doesn't support running that module instance. The other potential cause is if a module isn't available for a particular target and the user tries to use that combination.
 - Print and store `trace` field from a run properly when it's either canceled or failed
+- Add warning about nnxtb integration being experimental to docs
 - Remove AI-generated mismatches page from docs
 
 ## 6.10.1
diff --git a/docs/nnxtb/overview.md b/docs/nnxtb/overview.md
index edd60a15..3bf23f1c 100644
--- a/docs/nnxtb/overview.md
+++ b/docs/nnxtb/overview.md
@@ -1,5 +1,7 @@
 # Overview | [Use NN-xTB](https://nnxtb.qdx.co)
 
+> ⚠️ **Experimental Integration** — The NN-xTB integration with RushPy is currently experimental. You will likely encounter errors or unexpected behaviour.
+
 NN-xTB (Neural Network extended Tight Binding) is a GPU-accelerated semi-empirical quantum chemistry method built by QDX. It reparameterizes the GFN2-xTB Hamiltonian with a neural network to approach DFT-level accuracy while retaining the interpretability and speed of tight-binding methods.
 
 NN-xTB fills the gap between fast but approximate classical force fields and accurate but expensive *ab initio* quantum chemistry. It is designed for workflows where you need quantum-level electronic structure information — energies, forces, vibrational frequencies — but cannot afford the cost of full DFT or wavefunction methods across large numbers of structures.