diff --git a/nevergrad/functions/gym/multigym.py b/nevergrad/functions/gym/multigym.py
index 6e0f03ece..b50796ee9 100644
--- a/nevergrad/functions/gym/multigym.py
+++ b/nevergrad/functions/gym/multigym.py
@@ -668,6 +668,7 @@ def discretize(self, a, env):
                     tmp_env = copy.deepcopy(env)
                 _, r, _, _ = tmp_env.step(action)
                 a[i] += self.greedy_coefficient * r
+        a = np.nan_to_num(a, copy=False, nan=-1e20, posinf=1e20, neginf=-1e20)
         probabilities = np.exp(a - max(a))
         probabilities = probabilities / sum(probabilities)
         assert sum(probabilities) <= 1.0 + 1e-7, f"{probabilities} with greediness {self.greedy_coefficient}."
diff --git a/nevergrad/functions/gym/test_multigym.py b/nevergrad/functions/gym/test_multigym.py
index 14a22d7be..4d2f09867 100644
--- a/nevergrad/functions/gym/test_multigym.py
+++ b/nevergrad/functions/gym/test_multigym.py
@@ -56,6 +56,9 @@ def test_sparse_cartpole() -> None:
 def test_run_multigym(name: str) -> None:
     if os.name == "nt" or np.random.randint(8) or "CubeCrash" in name:
         raise SkipTest("Skipping Windows and running only 1 out of 8")
+    if "ANM" in name:
+        raise SkipTest("We skip ANM6Easy and related problems.")
+
     func = multigym.GymMulti(randomized=False, neural_factor=None)
     x = np.zeros(func.dimension)
     value = func(x)