diff --git a/Lib/test/test_free_threading/test_itertools.py b/Lib/test/test_free_threading/test_itertools.py
index bb6047e8669475..4772c83f598389 100644
--- a/Lib/test/test_free_threading/test_itertools.py
+++ b/Lib/test/test_free_threading/test_itertools.py
@@ -1,5 +1,5 @@
 import unittest
-from itertools import batched, chain, combinations_with_replacement, cycle, permutations
+from itertools import batched, chain, combinations_with_replacement, cycle, islice, permutations
 from test.support import threading_helper
 
 
@@ -48,6 +48,13 @@ def test_combinations_with_replacement(self):
             it = combinations_with_replacement(tuple(range(2)), 2)
             threading_helper.run_concurrently(work_iterator, nthreads=6, args=[it])
 
+    @threading_helper.reap_threads
+    def test_islice(self):
+        number_of_iterations = 6
+        for _ in range(number_of_iterations):
+            it = islice(tuple(range(10)), 1, 8, 2)
+            threading_helper.run_concurrently(work_iterator, nthreads=10, args=[it])
+
     @threading_helper.reap_threads
     def test_permutations(self):
         number_of_iterations = 6
diff --git a/Misc/NEWS.d/next/Library/2026-02-05-20-42-00.gh-issue-123471.Uj-Eyr.rst b/Misc/NEWS.d/next/Library/2026-02-05-20-42-00.gh-issue-123471.Uj-Eyr.rst
new file mode 100644
index 00000000000000..52bc301b439943
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2026-02-05-20-42-00.gh-issue-123471.Uj-Eyr.rst
@@ -0,0 +1 @@
+Make concurrent iterations over :class:`itertools.islice` safe under free-threading.
diff --git a/Modules/itertoolsmodule.c b/Modules/itertoolsmodule.c
index 7e73f76bc20b58..411c4f840962c8 100644
--- a/Modules/itertoolsmodule.c
+++ b/Modules/itertoolsmodule.c
@@ -6,6 +6,7 @@
 #include "pycore_moduleobject.h"      // _PyModule_GetState()
 #include "pycore_typeobject.h"        // _PyType_GetModuleState()
 #include "pycore_object.h"            // _PyObject_GC_TRACK()
+#include "pycore_pyatomic_ft_wrappers.h" // FT_ATOMIC_STORE_SSIZE_RELAXED()
 #include "pycore_tuple.h"             // _PyTuple_ITEMS()
 
 #include <stddef.h>                   // offsetof()
@@ -1617,36 +1618,44 @@ islice_next(PyObject *op)
     PyObject *item;
     PyObject *it = lz->it;
     Py_ssize_t stop = lz->stop;
-    Py_ssize_t oldnext;
     PyObject *(*iternext)(PyObject *);
 
-    if (it == NULL)
+    Py_ssize_t step = FT_ATOMIC_LOAD_SSIZE_RELAXED(lz->step);
+    if (step < 0)
         return NULL;
 
+    Py_ssize_t cnt = FT_ATOMIC_LOAD_SSIZE_RELAXED(lz->cnt);
+    Py_ssize_t oldnext = FT_ATOMIC_LOAD_SSIZE_RELAXED(lz->next);
     iternext = *Py_TYPE(it)->tp_iternext;
-    while (lz->cnt < lz->next) {
+    while (cnt < oldnext) {
         item = iternext(it);
         if (item == NULL)
             goto empty;
         Py_DECREF(item);
-        lz->cnt++;
+        cnt++;
     }
-    if (stop != -1 && lz->cnt >= stop)
+    if (stop != -1 && cnt >= stop)
         goto empty;
     item = iternext(it);
     if (item == NULL)
         goto empty;
-    lz->cnt++;
-    oldnext = lz->next;
+    cnt++;
+    FT_ATOMIC_STORE_SSIZE_RELAXED(lz->cnt, cnt);
+
     /* The (size_t) cast below avoids the danger of undefined
        behaviour from signed integer overflow. */
-    lz->next += (size_t)lz->step;
-    if (lz->next < oldnext || (stop != -1 && lz->next > stop))
-        lz->next = stop;
+    Py_ssize_t new_next = oldnext + (size_t)step;
+    FT_ATOMIC_STORE_SSIZE_RELAXED(lz->next, new_next);
+    if (new_next < oldnext || (stop != -1 && new_next > stop)) {
+        FT_ATOMIC_STORE_SSIZE_RELAXED(lz->next, stop);
+    }
     return item;
 
 empty:
+    FT_ATOMIC_STORE_SSIZE_RELAXED(lz->step, -1);
+#ifndef PY_GIL_DISABLED
     Py_CLEAR(lz->it);
+#endif
     return NULL;
 }
 
@@ -3555,7 +3564,7 @@ count_next(PyObject *op)
     PyObject *returned;
     Py_ssize_t cnt;
 
-    cnt = _Py_atomic_load_ssize_relaxed(&lz->cnt);
+    cnt = FT_ATOMIC_LOAD_SSIZE_RELAXED(lz->cnt);
     for (;;) {
         if (cnt == PY_SSIZE_T_MAX) {
             Py_BEGIN_CRITICAL_SECTION(lz);