diff --git a/.github/workflows/format-python.yml b/.github/workflows/format-python.yml
index 24b3127e..565c01d2 100644
--- a/.github/workflows/format-python.yml
+++ b/.github/workflows/format-python.yml
@@ -1,16 +1,20 @@
 name: Python Formatting Check
 
 on:
-  pull_request:
+  pull_request_target:
     types: [opened, synchronize, reopened]
 
 jobs:
   python-black-check:
     runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
 
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
 
       - name: Set up Python
         uses: actions/setup-python@v5
diff --git a/.github/workflows/format.yml b/.github/workflows/format.yml
index 54d64135..151a2c06 100644
--- a/.github/workflows/format.yml
+++ b/.github/workflows/format.yml
@@ -1,16 +1,20 @@
 name: Clang-Format Lint Check
 
 on:
-  pull_request:
+  pull_request_target:
     types: [opened, synchronize, reopened]
 
 jobs:
   clang-format-check:
     runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
 
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
 
       - name: Install clang-format
         run: sudo apt-get update && sudo apt-get install -y clang-format-19
diff --git a/include/memory/container.h b/include/memory/container.h
index 4c5e6a2e..0681e0e0 100644
--- a/include/memory/container.h
+++ b/include/memory/container.h
@@ -23,6 +23,20 @@ concept numeric = std::is_arithmetic_v<T>;
  */
 template <numeric T> struct device_span : public std::span<T> {
   device dev;
+  // Return raw pointers from begin()/end() instead of std::span's iterators.
+  //
+  // In libstdc++ 15 the tag type of std::span's iterator
+  // (std::span<T>::__iter_tag) is a private member. When a span iterator is
+  // handed to a thrust/CUB device algorithm (e.g. thrust::transform in
+  // MobilityInterface::sqrtMdotW), nvcc generates host stub code that names
+  // that private type, which GCC 15 rejects ("__iter_tag is private within
+  // this context"). This surfaces with CUDA >= 13 (whose bundled thrust/CUB
+  // routes span iterators through this path) built against GCC 15.
+  //
+  // Handing out raw pointers keeps device_span a valid contiguous range while
+  // ensuring the private tag is never instantiated.
+  constexpr T *begin() const noexcept { return this->data(); }
+  constexpr T *end() const noexcept { return this->data() + this->size(); }
   device_span(std::span<T> data, device dev) : std::span<T>(data), dev(dev) {}
   device_span() : std::span<T>(), dev(device::unknown) {}
   template <class Allocator>