One correction is noted, which does not affect the results or limits reported. The acceptance times efficiency numbers quoted for the nonresonant signal in the Letter (but not those used in the analysis) did not take into account the effect of the M AD G RAPH 5-aMC@NLO Monte Carlo event weights and are therefore incorrect. In the paragraph discussing the final fit and limit-setting procedure, the corrected sentence should read as follows: The acceptance times efficiency for the NR HH signal is 3.2% (1.9%) in the combined SLT and LTT τlep τhad (τhad τhad) channel over the full BDT distribution, decreasing to 1.8% (1.4%) for the two most sensitive BDT bins. © 2019 CERN for the ATLAS Collaboration.