There are always assumptions that you can play with. For example, the no-cloning theorem says that the transformation $$|\psi~0\rangle \mapsto |\psi~\psi\rangle$$is necessarily nonunitary and can never work with QM's rules of linearity; the proof is just that the state $$A|00\rangle + B|10\rangle \mapsto A|00\rangle + B|11\rangle$$works out to be an entangled state on the right-hand side, not a product of two coherent states.
So, one assumption is that quantum mechanics is right; you're not going to be able to do much with the reverse. But another assumption is that we're trying to clone the state from A to B rather than just trying to clone the state. In fact you can clone a state in general; just perform all of the experimental procedure that generated qubit 1 on qubit 2.
Similarly, no-communication is a super-robust principle which says that given any density matrix $\rho$ describing two entangled subsystems $A$ and $B$, there is a "reduced" density matrix $\rho_A = \operatorname{Tr}_B \rho$ which describes expectations for all observables of the form $\hat O_A \otimes \hat I_B$ as $\operatorname{Tr}\hat O_A \rho_A;$ we observe that no Hamiltonian of the form $\hat I_A \otimes \hat H_B$ can modify this because it leads to a unitary operator $I_A \otimes U_B$ constructing $$\rho_A' = \operatorname{Tr}_B (I_A \otimes U_B)~\rho~(I_A \otimes U_B)^\dagger = \operatorname{Tr}_B\rho = \rho_A.$$This proves that nothing which B does fundamentally alters what A sees in isolation.
But of course you can then cheat this system by comparing their measurements, especially by ignoring some of them, at which point it really does seem like B is affecting A and sending messages and so forth. That's how the quantum eraser experiment works: there is a "coincidence counter" which quietly erases a bunch of A's contributed data and then goes, "hey, B has an effect on A, even if it has to go back in time to have that effect!"
One of the most productive sets of publications dealt with the no-go theorems about quantum cryptography: given this protocol, it is not possible to hack the system; these two have for sure a private, shared set of bits. The assumption that these publications attacked were, the quantum hardware we're using is ideal. It's not! There's fudge factors, error tolerances, undefined behavior when you blast it with a high-power laser, and with a lot of these real-world properties can make your system look 100% secure when it is compromised.