Let $\Omega = \Omega_1 \cup \Omega_2$, where $\{ \Omega_i \}$ represents two elements that share an interface $\Gamma = \Omega_1 \cap \Omega_2$. Assume that $u \in H^2(\Omega)$. Multiplying $\Delta u$ by a smooth test function $v \in C^\infty$, we then can integrate by parts
\begin{equation} \label{eq:omega} \tag{1}
\int_\Omega (\Delta u) v \, dx = -\int_\Omega \nabla u \cdot \nabla v \, dx + \int_{\partial \Omega} v \nabla u \cdot \boldsymbol{n} \, ds
\end{equation}
where $\Delta u$ and $\nabla u$ are interpreted in the sense of weak derivatives.
The same also holds over each element $\Omega_i$,
\begin{equation} \label{eq:omega-i} \tag{2}
\int_{\Omega_i} (\Delta u) v \, dx = -\int_{\Omega_i} \nabla u \cdot \nabla v \, dx + \int_{\partial \Omega_i} v \nabla u \cdot \boldsymbol n_i \, ds,
\end{equation}
where $\boldsymbol n_i$ is the normal vector facing outwards from $\Omega_i$.
Summing \eqref{eq:omega-i} over $i=1,2$ we obtain
\begin{equation} \label{eq:sum} \tag{3}
\int_{\Omega} (\Delta u) v \, dx = -\int_{\Omega} \nabla u \cdot \nabla v \, dx + \int_{\partial \Omega} v \nabla u \cdot \boldsymbol{n} \, ds + \int_\Gamma v [ \nabla u ] \, ds.
\end{equation}
We see that in order for both \eqref{eq:omega} and \eqref{eq:sum} to hold, we must have that $\int_\Gamma v [ \nabla u ] \, ds = 0$, and since this holds for all test functions $v \in C^\infty$, we have that $[ \nabla u ] = 0$.
In the above, I am using the (fairly standard) notation that for a vector-valued quantity $\boldsymbol \varphi$, the jump is defined by $[ \boldsymbol \varphi ] = \boldsymbol \varphi_1 \cdot \boldsymbol n_1 + \boldsymbol \varphi_2 \cdot \boldsymbol n_2$, where $\boldsymbol \varphi_i$ is the trace from within $\Omega_i$.