diff --git a/ext/TensorKitEnzymeExt/indexmanipulations.jl b/ext/TensorKitEnzymeExt/indexmanipulations.jl
index 3a54a3230..e0c2968f7 100644
--- a/ext/TensorKitEnzymeExt/indexmanipulations.jl
+++ b/ext/TensorKitEnzymeExt/indexmanipulations.jl
@@ -18,7 +18,10 @@ for transform in (:permute, :transpose)
         # and store that instead of repeating the permutation in the pullback each time.
         # effectively, we replace `add_permute` by `add ∘ permute`.
         Ap = if !isa(α, Const)
-            Ap = $transform(A.val, p.val)
+            # this may just alias A if the permutation is an identity or the space
+            # is Trivial and has_shared_permute is true.
+            Ap_needs_copy = (p.val == (TensorKit.codomainind(A.val), TensorKit.domainind(A.val)) || TensorKit.has_shared_permute(A.val, p.val))
+            Ap = $transform(A.val, p.val; copy = Ap_needs_copy)
             add!(C.val, Ap, α.val, β.val)
             Ap
         else
diff --git a/test/enzyme/inv.jl b/test/enzyme/inv.jl
index 35bd074fb..20475e4e9 100644
--- a/test/enzyme/inv.jl
+++ b/test/enzyme/inv.jl
@@ -21,12 +21,9 @@ TDs = is_ci ? (Duplicated,) : (Const, Duplicated)
                 D2 = randn(T, V[1] ⊗ V[2] ← V[1] ⊗ V[2])
                 EnzymeTestUtils.test_reverse(inv, TD, (D2, TD); atol, rtol)
                 EnzymeTestUtils.test_forward(inv, TD, (D2, TD); atol, rtol)
-                # see https://github.com/QuantumKitHub/TensorKit.jl/issues/457
-                @static if VERSION ≥ v"1.11.0-rc"
-                    D3 = randn(T, V[1] ⊗ V[2] ⊗ V[3] ← V[1] ⊗ V[2] ⊗ V[3])
-                    EnzymeTestUtils.test_reverse(inv, TD, (D3, TD); atol, rtol)
-                    EnzymeTestUtils.test_forward(inv, TD, (D3, TD); atol, rtol)
-                end
+                D3 = randn(T, V[1] ⊗ V[2] ⊗ V[3] ← V[1] ⊗ V[2] ⊗ V[3])
+                EnzymeTestUtils.test_reverse(inv, TD, (D3, TD); atol, rtol)
+                EnzymeTestUtils.test_forward(inv, TD, (D3, TD); atol, rtol)
             end
         end
     end
diff --git a/test/enzyme/norm.jl b/test/enzyme/norm.jl
index 504e51884..d12dccc1a 100644
--- a/test/enzyme/norm.jl
+++ b/test/enzyme/norm.jl
@@ -14,12 +14,7 @@ fRTs = is_ci ? (Duplicated,) : (Const, Duplicated)
     @timedtestset "$(TensorKit.type_repr(sectortype(eltype(V)))) ($T), TC $TC" for V in spacelist, T in eltypes, TC in (Const, Duplicated)
         atol = default_tol(T)
         rtol = default_tol(T)
-        # see https://github.com/QuantumKitHub/TensorKit.jl/issues/457
-        @static if VERSION < v"1.11.0-rc"
-            C = randn(T, V[1] ⊗ V[2] ← (V[4] ⊗ V[5])')
-        else
-            C = randn(T, V[1] ⊗ V[2] ← (V[3] ⊗ V[4] ⊗ V[5])')
-        end
+        C = randn(T, V[1] ⊗ V[2] ← (V[3] ⊗ V[4] ⊗ V[5])')
         for RT in rRTs
             EnzymeTestUtils.test_reverse(norm, RT, (C, TC), (2, Const); atol, rtol)
             EnzymeTestUtils.test_reverse(norm, RT, (C', TC), (2, Const); atol, rtol)
diff --git a/test/enzyme/permute.jl b/test/enzyme/permute.jl
index 4e898f3a3..ffe60f89f 100644
--- a/test/enzyme/permute.jl
+++ b/test/enzyme/permute.jl
@@ -24,13 +24,11 @@ fTβs = is_ci ? (Duplicated,) : (Duplicated, Const)
         β = randn(T)
         p = randindextuple(numind(A))
         C = randn!(permute(A, p))
-        @static if VERSION ≥ v"1.11.0-rc"
-            for Tα in rTαs, Tβ in rTβs
-                EnzymeTestUtils.test_reverse(TensorKit.permute!, Duplicated, (C, Duplicated), (A, Duplicated), (p, Const), (α, Tα), (β, Tβ); atol, rtol, testset_name = "permute! reverse Tα: $Tα, Tβ: $Tβ")
-            end
-            for Tα in fTαs, Tβ in fTβs
-                EnzymeTestUtils.test_forward(TensorKit.permute!, Duplicated, (C, Duplicated), (A, Duplicated), (p, Const), (α, Tα), (β, Tβ); atol, rtol, testset_name = "permute! forward Tα: $Tα, Tβ: $Tβ")
-            end
+        for Tα in rTαs, Tβ in rTβs
+            EnzymeTestUtils.test_reverse(TensorKit.permute!, Duplicated, (C, Duplicated), (A, Duplicated), (p, Const), (α, Tα), (β, Tβ); atol, rtol, testset_name = "permute! reverse Tα: $Tα, Tβ: $Tβ")
+        end
+        for Tα in fTαs, Tβ in fTβs
+            EnzymeTestUtils.test_forward(TensorKit.permute!, Duplicated, (C, Duplicated), (A, Duplicated), (p, Const), (α, Tα), (β, Tβ); atol, rtol, testset_name = "permute! forward Tα: $Tα, Tβ: $Tβ")
         end
     end
 end
diff --git a/test/enzyme/tr.jl b/test/enzyme/tr.jl
index f42e02ee1..1ba0c2df7 100644
--- a/test/enzyme/tr.jl
+++ b/test/enzyme/tr.jl
@@ -22,18 +22,12 @@ TDs = is_ci ? (Duplicated,) : (Const, Duplicated)
         @testset "tr reverse: RT $RT, TD $TD" for RT in rRTs, TD in TDs
             EnzymeTestUtils.test_reverse(tr, RT, (D1, TD); atol, rtol)
             EnzymeTestUtils.test_reverse(tr, RT, (D2, TD); atol, rtol)
-            # see https://github.com/QuantumKitHub/TensorKit.jl/issues/457
-            @static if VERSION ≥ v"1.11.0-rc"
-                EnzymeTestUtils.test_reverse(tr, RT, (D3, TD); atol, rtol)
-            end
+            EnzymeTestUtils.test_reverse(tr, RT, (D3, TD); atol, rtol)
         end
         @testset "tr forward: RT $RT, TD $TD" for RT in fRTs, TD in TDs
             EnzymeTestUtils.test_forward(tr, RT, (D1, TD); atol, rtol)
             EnzymeTestUtils.test_forward(tr, RT, (D2, TD); atol, rtol)
-            # see https://github.com/QuantumKitHub/TensorKit.jl/issues/457
-            @static if VERSION ≥ v"1.11.0-rc"
-                EnzymeTestUtils.test_forward(tr, RT, (D3, TD); atol, rtol)
-            end
+            EnzymeTestUtils.test_forward(tr, RT, (D3, TD); atol, rtol)
         end
     end
 end
diff --git a/test/runtests.jl b/test/runtests.jl
index ae8446f2a..de3aed430 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -41,13 +41,14 @@ args = parse_args(ARGS; custom = ["fast"])
 # --fast: skip AD tests and inject fast_tests=true into each worker sandbox
 fast = !isnothing(args.custom["fast"])
 
-# Enzyme workers end up needing ~4 GB, but the default job count assumes ~2 GB per worker.
-# Cap the number of jobs by increasing the per-worker memory budget unless --jobs was given.
+# Enzyme workers peak at ~4.5 GB, but the default job count
+# assumes ~2 GB per worker. Cap the number of jobs by increasing the per-worker memory budget
+# unless --jobs was given; with 4 GB, 16 GB CI runners would run out of memory and be killed.
 selected = copy(testsuite)
 ParallelTestRunner.filter_tests!(selected, args)
 has_enzyme = any(startswith("enzyme"), keys(selected))
 if isnothing(args.jobs) && has_enzyme
-    njobs = clamp(Int(Sys.free_memory() ÷ (4 * Int64(2)^30)), 1, Sys.CPU_THREADS)
+    njobs = clamp(Int(Sys.free_memory() ÷ (5 * Int64(2)^30)), 1, Sys.CPU_THREADS)
     # On Windows with Julia 1.10, memory pressure makes recycled workers too slow to start:
     # Malt gives up if a new worker does not connect within 30 s, which aborts the whole run.
     Sys.iswindows() && VERSION < v"1.11" && (njobs = min(njobs, 2))