We generalise Merzlyakov's theorem about the first-order theory of nonabelian free groups to all acylindrically hyperbolic groups. As a corollary, we deduce that if G is an acylindrically hyperbolic group and E(G) denotes the unique maximal finite normal subgroup of G, then G and the HNN extension G * E(G) , which is simply the free product G * Z when E(G) is trivial, have the same ∀∃-theory. As a consequence, we prove the following conjecture, formulated by Casals-Ruiz, Garreta and de la Nuez González: acylindrically hyperbolic groups have trivial positive theory. In particular, one recovers a result proved by Bestvina, Bromberg and Fujiwara, stating that, with only the obvious exceptions, verbal subgroups of acylindrically hyperbolic groups have infinite width.(2) Is G elementarily embedded into G * E(G) ?As mentioned before, Sela proved that the answer to both of these questions is 'Yes' under the stronger assumption that G is a torsion-free non-elementary hyperbolic group or a non-trivial and non-dihedral free product. In all other cases, the answer is not known.Moreover, let us note that we do not know of any example of a finitely generated group G that is not acylindrically hyperbolic and that has the same first-order theory, or even the same ∀∃-theory, as G * Z. The question of the existence of such a group is closely related to that of the preservation of acylindrical hyperbolicity under elementary equivalence among finitely generated groups (see Section 10 for further comments). It is worth mentioning the following corollary of Theorem 1.3 (see Proposition 10.5).Corollary 1.7. Let G be an acylindrically hyperbolic group, and let H be a group that admits a non-trivial splitting over a virtually abelian group. Suppose that G and H are elementarily equivalent (or simply that they have the same ∃∀∃-theory). Then the group H is acylindrically hyperbolic.Remark 1.8. As a consequence, if there exists a group G that is not acylindrically hyperbolic and such that G and G * Z are elementarily equivalent, then all non-trivial splittings of G (if they exist) have sufficiently complicated edge groups. For instance, if G is a generalized Baumslag-Solitar group, then G and G * Z are not elementarily equivalent.