We study the potential of Bayesian Neural Networks (BNNs) to detect new physics in the dark matter power spectrum, concentrating here on evolving dark energy and modifications to General Relativity. After introducing a new technique to quantify classification uncertainty in BNNs, we train two BNNs on mock matter power spectra produced using the publicly available code ReACT in the k-range (0.01 − 2.5) hMpc −1 and redshift bins (0.1, 0.478, 0.783, 1.5) with Euclid-like noise. The first network classifies spectra into five labels including ΛCDM, f (R), wCDM, Dvali-Gabadaze-Porrati (DGP) gravity and a "random" class whereas the second is trained to distinguish ΛCDM from non-ΛCDM. Both networks achieve a comparable training, validation and test accuracy of ∼ 95%. Each network is also capable of detecting deviations from ΛCDM that were not included in the training set, demonstrated with spectra generated using the growth-index γ. We then quantify the constraining power of each network by computing the smallest deviation from ΛCDM such that the noise-averaged non-ΛCDM classification probability is at least 2σ, finding these bounds to be f R0 10 −7 , Ω rc 10 −2 , −1.05 w 0 0.95, −0.2 w a 0.2, 0.52 γ 0.59. The bounds on f (R) can be improved by training a specialist network to distinguish solely between ΛCDM and f (R) power spectra which can detect a non-zero f R0 at O 10 −8 with a confidence > 2σ. We expect that further developments, such as the inclusion of smaller length scales or additional extensions to ΛCDM, will only improve the potential of BNNs to detect new physics using cosmological datasets. Alongside this paper we publish the publicly available code Bayesian Cosmological Network (BaCoN) which can be accessed at the github repository https://github.com/Mik3M4n/BaCoN with the training and test data available at https://doi.org/10.5281/zenodo.4309918.