Non-equilibrium systems in steady states are commonly described by generalized statistical mechanical frameworks such as superstatistics, which assumes that the inverse temperature
β
=
1
/
(
k
B
T
)
is an unknown quantity having some pre-established statistical distribution. The uncertainty in β is usually understood as the fluctuation of a physical observable, however, it has been previously proved (Davis and Gutiérrez 2018 Physica A 505 864–70) that β in a superstatistical model cannot be associated to an observable function
B
(
Γ
)
of the microstates Γ. In this work, we provide an information-theoretical interpretation of this theorem by introducing a new quantity
D
, the mutual information between β and Γ. Our results show that
D
is also a measure of departure from canonical equilibrium, and reveal a minimum, non-zero uncertainty about β given Γ for every non-canonical superstatistical ensemble. The behavior of
D
is illustrated in the case of a collisionless plasma described by kappa distributions, revealing the precise sense in which the spectral index κ can be understood as a measure of distance from equilibrium.