Recent experiments have put the claim to test Overview https://www.nature.com/articles/nature.2016.20282 The actual work https://www.nature.com/articles/ncomms12172 > Here we report that humans can detect a single-photon incident on the cornea with a probability significantly above chance.
Interesting study, though if you look at the results 'significantly above chance' looks to be 0.60 +/- .05, which is indeed better than 0.50 but not what anyone could call reliable detection. I wonder if trained owls could detect single photons, or if their night vision is based on just having much larger lenses that collect more light? It seems that all rods in retinas are activated by single-photon-absorption, it's…
"Significantly" in statistics does not mean "reliably", but "confidently".