diff --git a/ext/TensorKitEnzymeExt/TensorKitEnzymeExt.jl b/ext/TensorKitEnzymeExt/TensorKitEnzymeExt.jl
index 6b3e9c283..cadae496a 100644
--- a/ext/TensorKitEnzymeExt/TensorKitEnzymeExt.jl
+++ b/ext/TensorKitEnzymeExt/TensorKitEnzymeExt.jl
@@ -3,7 +3,7 @@ module TensorKitEnzymeExt
 using Enzyme
 using TensorKit
 import TensorKit as TK
-using TensorKit: subblock
+using TensorKit: subblock, block
 using VectorInterface
 using TensorOperations: TensorOperations, IndexTuple, Index2Tuple, linearize
 import TensorOperations as TO
diff --git a/ext/TensorKitEnzymeExt/indexmanipulations.jl b/ext/TensorKitEnzymeExt/indexmanipulations.jl
index f772348e5..0adbb0600 100644
--- a/ext/TensorKitEnzymeExt/indexmanipulations.jl
+++ b/ext/TensorKitEnzymeExt/indexmanipulations.jl
@@ -148,92 +148,3 @@ function EnzymeRules.reverse(
     end
     return (nothing, nothing)
 end
-
-for insertunit in (:insertleftunit, :insertrightunit)
-    @eval begin
-        function EnzymeRules.augmented_primal(
-                config::EnzymeRules.RevConfigWidth{1},
-                func::Const{typeof($insertunit)},
-                ::Type{RT},
-                tsrc::Annotation{<:AbstractTensorMap},
-                ival::Const{<:Val};
-                kwargs...
-            ) where {RT}
-            if tsrc.val isa TensorMap && !get(kwargs, :copy, false) && !isa(tsrc, Const)
-                tsrc_cache = tsrc.val
-                tdst = $insertunit(tsrc.val, ival.val; kwargs...)
-                Δtdst = $insertunit(tsrc.dval, ival.val; kwargs...)
-            else
-                tsrc_cache = nothing
-                tdst = $insertunit(tsrc.val, ival.val; kwargs...)
-                Δtdst = make_zero(tdst)
-            end
-            primal = EnzymeRules.needs_primal(config) ? tdst : nothing
-            shadow = EnzymeRules.needs_shadow(config) ? Δtdst : nothing
-            cache = (tsrc_cache, tdst, Δtdst)
-            return EnzymeRules.AugmentedReturn(primal, shadow, cache)
-        end
-        function EnzymeRules.reverse(
-                config::EnzymeRules.RevConfigWidth{1},
-                func::Const{typeof($insertunit)},
-                ::Type{RT},
-                cache,
-                tsrc::Annotation{<:AbstractTensorMap},
-                ival::Const{<:Val};
-                kwargs...
-            ) where {RT}
-            tsrc_cache, tdst, Δtdst = cache
-            # note: since data is already shared for <:TensorMap, don't have to do anything here!
-            if isnothing(tsrc_cache) && !isa(tsrc, Const)
-                for (c, b) in blocks(Δtdst)
-                    add!(block(tsrc.dval, c), b)
-                end
-            end
-            return (nothing, nothing)
-        end
-    end
-end
-
-function EnzymeRules.augmented_primal(
-        config::EnzymeRules.RevConfigWidth{1},
-        func::Const{typeof(removeunit)},
-        ::Type{RT},
-        tsrc::Annotation{<:AbstractTensorMap},
-        ival::Const{<:Val};
-        kwargs...
-    ) where {RT}
-    # tdst shares data with tsrc if <:TensorMap & copy=false, in this case we have to deal with correctly
-    # sharing address spaces
-    if tsrc.val isa TensorMap && !get(kwargs, :copy, false) && !isa(tsrc, Const)
-        tsrc_cache = tsrc.val
-        tdst = removeunit(tsrc.val, ival.val; kwargs...)
-        Δtdst = removeunit(tsrc.dval, ival.val)
-    else
-        tsrc_cache = nothing
-        tdst = removeunit(tsrc.val, ival.val; kwargs...)
-        Δtdst = make_zero(tdst)
-    end
-    primal = EnzymeRules.needs_primal(config) ? tdst : nothing
-    shadow = EnzymeRules.needs_shadow(config) ? Δtdst : nothing
-    cache = (tsrc_cache, tdst, Δtdst)
-    return EnzymeRules.AugmentedReturn(primal, shadow, cache)
-end
-
-function EnzymeRules.reverse(
-        config::EnzymeRules.RevConfigWidth{1},
-        func::Const{typeof(removeunit)},
-        ::Type{RT},
-        cache,
-        tsrc::Annotation{<:AbstractTensorMap},
-        ival::Const{<:Val};
-        kwargs...
-    ) where {RT}
-    tsrc_cache, tdst, Δtdst = cache
-    # note: since data for <: TensorMap is already shared, don't have to do anything here!
-    if isnothing(tsrc_cache) && !isa(tsrc, Const)
-        for (c, b) in blocks(Δtdst)
-            add!(block(tsrc.dval, c), b)
-        end
-    end
-    return (nothing, nothing)
-end
diff --git a/ext/TensorKitEnzymeExt/utility.jl b/ext/TensorKitEnzymeExt/utility.jl
index a364c5daf..dd99d96e8 100644
--- a/ext/TensorKitEnzymeExt/utility.jl
+++ b/ext/TensorKitEnzymeExt/utility.jl
@@ -45,10 +45,35 @@ function EnzymeRules.reverse(
         t::Annotation{<:AbstractTensorMap},
         f::Annotation,
     ) where {RT}
-    dret = cache
-    if !isnothing(dret) && !isa(t, Const)
-        subblock(t.dval, f.val) .= dret
+    return (nothing, nothing)
+end
+
+function EnzymeRules.augmented_primal(
+        config::EnzymeRules.RevConfigWidth{1},
+        ::Const{typeof(block)},
+        ::Type{RT},
+        t::Annotation{<:AbstractTensorMap},
+        c::Annotation{<:Sector},
+    ) where {RT}
+    ret = EnzymeRules.needs_primal(config) ? block(t.val, c.val) : nothing
+    dret = if !isa(t, Const) && EnzymeRules.needs_shadow(config)
+        block(t.dval, c.val)
+    elseif EnzymeRules.needs_shadow(config)
+        Enzyme.make_zero(ret)
+    else
+        nothing
     end
+    return EnzymeRules.AugmentedReturn(ret, dret, dret)
+end
+
+function EnzymeRules.reverse(
+        config::EnzymeRules.RevConfigWidth{1},
+        ::Const{typeof(block)},
+        ::Type{RT},
+        cache,
+        t::Annotation{<:AbstractTensorMap},
+        c::Annotation{<:Sector},
+    ) where {RT}
     return (nothing, nothing)
 end
 
@@ -78,9 +103,38 @@ function EnzymeRules.forward(
     end
 end
 
+function EnzymeRules.forward(
+        config::EnzymeRules.FwdConfigWidth{1},
+        ::Const{typeof(block)},
+        ::Type{RT},
+        t::Annotation{<:AbstractTensorMap},
+        c::Annotation{<:Sector},
+    ) where {RT}
+    ret = EnzymeRules.needs_primal(config) ? block(t.val, c.val) : nothing
+    dret = if !isa(t, Const) && EnzymeRules.needs_shadow(config)
+        block(t.dval, c.val)
+    elseif EnzymeRules.needs_shadow(config)
+        Enzyme.make_zero(ret)
+    else
+        nothing
+    end
+    if EnzymeRules.needs_primal(config) && EnzymeRules.needs_shadow(config)
+        return Duplicated(ret, dret)
+    elseif EnzymeRules.needs_primal(config)
+        return ret
+    elseif EnzymeRules.needs_shadow(config)
+        return dret
+    else
+        return nothing
+    end
+end
+
 @inline EnzymeRules.inactive(::typeof(TensorKit.fsbraid), ::Any) = nothing
 @inline EnzymeRules.inactive(::typeof(TensorKit.fsbraid), ::Any, ::Any) = nothing
 @inline EnzymeRules.inactive(::typeof(TensorKit.artin_braid), ::Any, ::Any) = nothing
+@inline EnzymeRules.inactive(::typeof(TensorKit.insertleftunit), ::HomSpace, ::Any) = nothing
+@inline EnzymeRules.inactive(::typeof(TensorKit.insertrightunit), ::HomSpace, ::Any) = nothing
+@inline EnzymeRules.inactive(::typeof(TensorKit.removeunit), ::HomSpace, ::Any) = nothing
 @inline EnzymeRules.inactive(::typeof(TensorKit.sectorstructure), ::Any) = nothing
 @inline EnzymeRules.inactive(::typeof(TensorKit.degeneracystructure), ::Any) = nothing
 @inline EnzymeRules.inactive(::typeof(TensorKit.select), s::HomSpace, i::Index2Tuple) = nothing
diff --git a/test/enzyme-indexmanipulations-unit/insertunit.jl b/test/enzyme-indexmanipulations-unit/insertunit.jl
index 9513d5ee7..bd59f9d62 100644
--- a/test/enzyme-indexmanipulations-unit/insertunit.jl
+++ b/test/enzyme-indexmanipulations-unit/insertunit.jl
@@ -6,21 +6,33 @@ using Random
 spacelist = ad_spacelist(fast_tests)
 eltypes = (Float64, ComplexF64)
 
-if VERSION > v"1.11.0-rc" # https://github.com/QuantumKitHub/TensorKit.jl/issues/457
-    @timedtestset verbose = true "Enzyme - Index Manipulations (insertunit):" begin
-        @timedtestset verbose = true "$(TensorKit.type_repr(sectortype(eltype(V)))) ($T)" for V in spacelist, T in eltypes, TA in (Duplicated,)
-            atol = default_tol(T)
-            rtol = default_tol(T)
-            A = randn(T, V[1] ⊗ V[2] ← (V[3] ⊗ V[4] ⊗ V[5])')
-            @testset for insertunit in (insertleftunit, insertrightunit)
-                EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(1), Const); atol, rtol)
-                EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(4), Const); atol, rtol)
+is_ci = get(ENV, "CI", "false") == "true"
+
+@timedtestset verbose = true "Enzyme - Index Manipulations (insertunit):" begin
+    @timedtestset verbose = true "$(TensorKit.type_repr(sectortype(eltype(V)))) ($T)" for V in spacelist, T in eltypes, TA in (Duplicated,)
+        atol = default_tol(T)
+        rtol = default_tol(T)
+        A = randn(T, V[1] ⊗ V[2] ← (V[3] ⊗ V[4] ⊗ V[5])')
+        @testset for insertunit in (insertleftunit, insertrightunit)
+            EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(1), Const); atol, rtol)
+            EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(4), Const); atol, rtol)
+            if !is_ci
                 EnzymeTestUtils.test_reverse(insertunit, TA, (A', TA), (Val(2), Const); atol, rtol)
                 EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(1), Const); atol, rtol, fkwargs = (copy = false,))
                 EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(2), Const); atol, rtol, fkwargs = (copy = true,))
                 EnzymeTestUtils.test_reverse(insertunit, TA, (A, TA), (Val(3), Const); atol, rtol, fkwargs = (copy = false, dual = true, conj = true))
                 EnzymeTestUtils.test_reverse(insertunit, TA, (A', TA), (Val(3), Const); atol, rtol, fkwargs = (copy = false, dual = true, conj = true))
             end
+
+            EnzymeTestUtils.test_forward(insertunit, TA, (A, TA), (Val(1), Const); atol, rtol)
+            EnzymeTestUtils.test_forward(insertunit, TA, (A, TA), (Val(4), Const); atol, rtol)
+            if !is_ci
+                EnzymeTestUtils.test_forward(insertunit, TA, (A', TA), (Val(2), Const); atol, rtol)
+                EnzymeTestUtils.test_forward(insertunit, TA, (A, TA), (Val(1), Const); atol, rtol, fkwargs = (copy = false,))
+                EnzymeTestUtils.test_forward(insertunit, TA, (A, TA), (Val(2), Const); atol, rtol, fkwargs = (copy = true,))
+                EnzymeTestUtils.test_forward(insertunit, TA, (A, TA), (Val(3), Const); atol, rtol, fkwargs = (copy = false, dual = true, conj = true))
+                EnzymeTestUtils.test_forward(insertunit, TA, (A', TA), (Val(3), Const); atol, rtol, fkwargs = (copy = false, dual = true, conj = true))
+            end
         end
     end
 end
diff --git a/test/enzyme-indexmanipulations-unit/removeunit.jl b/test/enzyme-indexmanipulations-unit/removeunit.jl
index 2cd632a90..a5b42d0bb 100644
--- a/test/enzyme-indexmanipulations-unit/removeunit.jl
+++ b/test/enzyme-indexmanipulations-unit/removeunit.jl
@@ -6,17 +6,18 @@ using Random
 spacelist = ad_spacelist(fast_tests)
 eltypes = (Float64, ComplexF64)
 
-if VERSION > v"1.11.0-rc" # https://github.com/QuantumKitHub/TensorKit.jl/issues/457
-    @timedtestset verbose = true "Enzyme - Index Manipulations (removeunit):" begin
-        @timedtestset verbose = true "$(TensorKit.type_repr(sectortype(eltype(V)))) ($T)" for V in spacelist, T in eltypes, TB in (Duplicated,)
-            atol = default_tol(T)
-            rtol = default_tol(T)
-            A = randn(T, V[1] ⊗ V[2] ← (V[3] ⊗ V[4] ⊗ V[5])')
-            for i in 1:2
-                B = insertleftunit(A, i; dual = rand(Bool))
-                EnzymeTestUtils.test_reverse(removeunit, TB, (B, TB), (Val(i), Const); atol, rtol, fkwargs = (copy = false,))
-                EnzymeTestUtils.test_reverse(removeunit, TB, (B, TB), (Val(i), Const); atol, rtol, fkwargs = (copy = true,))
-            end
+@timedtestset verbose = true "Enzyme - Index Manipulations (removeunit):" begin
+    @timedtestset verbose = true "$(TensorKit.type_repr(sectortype(eltype(V)))) ($T)" for V in spacelist, T in eltypes, TB in (Duplicated,)
+        atol = default_tol(T)
+        rtol = default_tol(T)
+        A = randn(T, V[1] ⊗ V[2] ← (V[3] ⊗ V[4] ⊗ V[5])')
+        for i in 1:2
+            B = insertleftunit(A, i; dual = rand(Bool))
+            EnzymeTestUtils.test_reverse(removeunit, TB, (B, TB), (Val(i), Const); atol, rtol, fkwargs = (copy = false,))
+            EnzymeTestUtils.test_reverse(removeunit, TB, (B, TB), (Val(i), Const); atol, rtol, fkwargs = (copy = true,))
+
+            EnzymeTestUtils.test_forward(removeunit, TB, (B, TB), (Val(i), Const); atol, rtol, fkwargs = (copy = false,))
+            EnzymeTestUtils.test_forward(removeunit, TB, (B, TB), (Val(i), Const); atol, rtol, fkwargs = (copy = true,))
         end
     end
 end