For discovering an effect the standard is indeed 5-sigma. But here we are discussing excluding an effect, namely deciding that a suggested effect isn't there. For this the standard is lower.