diff --git a/environment.yml b/environment.yml
index 7b21de87f..e87e31e6d 100644
--- a/environment.yml
+++ b/environment.yml
@@ -21,10 +21,10 @@ dependencies:
   - pip:
       # PyTorch via pip: the default Linux wheels are CUDA-enabled (cu124), whereas the
       # conda 'pytorch' channel yields a CPU-only build when this env is created on a
-      # machine with no NVIDIA driver (the VNN-COMP installer). Pinned to 2.5.1/cu124
-      # (needs driver >= 525) so the GPU build works on standard AWS run instances.
-      - torch
-      - torchvision
+      # machine with no NVIDIA driver (the VNN-COMP installer). Installing via pip instead
+      # of conda so the GPU build works on standard AWS run instances.
+      - torch>=2.8
+      - torchvision>=0.23
       # ONNX ecosystem - for VNNLib benchmarks
       - onnx2torch
       - onnxconverter-common