No, using XEX mode with $j = 0$ is not entirely safe. As noted in section 6 of the Rogaway (2004) paper (emphasis mine):
"Some added care is needed to address the security of XEX. Suppose, to be concrete, that we are looking at $\mathsf{XEX}[E,2^{\mathbf I}]$ and $\mathbf I=[0\mathop{..}2^{n-2}]$. Let the adversary ask a deciphering query with ciphertext $C=0^n$ and tweak $(0^n,0)$. If the adversary has a construction-based deciphering oracle then it will get a response of $M=\tilde D_K^{0^n}\,^0(0^n)=D_K(Δ)⊕Δ=D_K(\mathsf N)⊕\mathsf N=0^n⊕\mathsf N=\mathsf N$, where $\mathsf N = E_K(0^n) = Δ$. This allows the adversary to defeat the CCA-security. For example, enciphering $2M=2\mathsf N$ with a tweak of $(0^n,1)$ and enciphering $4M=4\mathsf N$ with a tweak of $(0^n,2)$ will give
identical results (if the adversary has the construction-based enciphering oracle). Corresponding to this attack we exclude any tweak $(N,i_1,\dotsc,i_k)$ for which $(i_1,\dotsc,i_k)$ is a representative of $1$—that is, any tweak $(N,i_1,\dotsc,i_k)$ for which $α_1^{i_1} \dotsb α_k^{i_k} = 1$. In particular, this condition excludes any tweak $(N,0,\dotsc,0)$."
However, using $j = 1$ is safe, and, indeed, is explicitly used as an example of a valid tweak by Rogaway e.g. in section 3:
"Tweaks that are not the increment of a prior tweak will also
arise, and they will typically look like $(N,1,0\dotsc,0)$. Constructions should be reasonably efficient in dealing with such tweaks."
Also, in their comments on the XTS mode, Liskov and Minematsu confirm that, indeed, $j=0$ is a bad choice for XEX (emphasis original):
"Note that $j = 0$ must be excluded, as $f(0, v) = v$ for any $v$, which implies $ρ = 1$. Moreover, if $j = 0$ was allowed, a simple attack based on this fact existed, as pointed out by [6] and [3]. Hence if XEX is used, one must be careful to avoid $j$ being $0$."
However (while generally criticizing the choice to use two keys for XTS), they also note that this requirement to avoid $j = 0$ only applies to the single-key XEX mode, and not to XTS, which uses separate keys for the tweak and encryption stages:
"Note that XTS does not require to avoid $j = 0$, as the offset function of XTS needs not be $(ϵ,γ,ρ)$-uniform, but only be $ϵ$-AXU if it is combined with URP. This difference is significant in security, but has no impact on effectiveness for practical applications."
