Let us consider an action $S[\phi,\partial\phi]$ which is classically invariant under a transformation group $G$. The associated Noether current $\mathcal{J}^\mu$ is classically conserved, namely $\partial_\mu \mathcal{J}^\mu=0$ holds as an operatorial identity.
The generating functional of this theory is
$$ Z[J] = \int [d\phi] \,e^{iS + i \int d^dx \,J(x)\phi(x)} $$
where we added an external source for $\phi(x)$.
If the theory is anomaly-free, we can derive the Ward-Identity for $\mathcal{J}^\mu$, which is
$$ \partial_\mu \langle \mathcal{J}^\mu(x)\rangle_J = J(x) \langle \delta \phi(x)\rangle_J\,,\qquad \qquad (1) $$
where $\langle ... \rangle_J$ means that the correlations must be computed with $J\neq 0$ and $\delta \phi$ is the variation of the field $\phi$ under $G$.
By differentiating Eq. (1) wrt $J(x_i)$ and then setting $J=0$, we get the usual Ward identities.
Usually Eq. (1) is the departing point to state that conserved currents do not acquire anomalous dimension. Indeed, the Ward-identity can be written roughly as
$$ \partial_\mu \langle \mathcal{J}^\mu(x)X(y)\rangle_{J=0} \propto \delta^d(x-y)\langle\delta X(y)\rangle $$ for some local product of fields $X(y)$. When the RHS is renormalized, the same is true for the LHS. Then, no renormalization of $\mathcal{J}^\mu$ is needed. This implies the dimension of $\mathcal{J}^\mu$ is fixed.
First question How does Eq.(1) change if the theory is anomalous? For example, put a conformally invariant theory on a curved background. We have a trace anomaly for the trace of the stress-energy momentum tensor, namely $\langle T^\mu_\mu\rangle \neq 0$. Consider as classically conserved current the dilatation current $D^\mu = x_\rho T^{\rho\mu}$. More precisely, If I know the matter content of the theory, I can see how the path-integral measure changes and then I can derive the anomalous Ward identity. What if instead I don't specify the matter content, but just the trace anomaly?
Second question If the theory is anomalous, is still true that classically conserved currents do not acquire anomalous dimension?