Harmful outcomes of psychological interventions are under-researched, including in mindfulness-based interventions (MBI) for psychosis. This systematic review summarizes reporting and prevalence of 8 harm indices (death, adverse events, hospitalisation, study drop out, noncompletion of therapy, side effects of therapy, symptom deterioration and crisis service use) in Randomised Controlled Trials (RCTs) of MBIs for psychosis. Meta-analyses of risk differences were also calculated for each harm index. The review included 39 studies, with a total n of 2684 participants across studies. The percentage of studies reporting on each index of harm, and the prevalence of harm, varied greatly across each index. 0% of studies reported on side effects of interventions compared to 92% of studies reporting on study dropout. Meta-analyses of risk differences (RD) found a higher risk of hospitalisation (RD (95% CI) = -0.136 (-0.23 to -0.05), p = 0.003) and crisis service use (RD (95% CI) = -0.160 (-0.299, -0.024), p = 0.02) in control arms compared to intervention arms, and no significant difference in adverse events, death, symptom deterioration, noncompletion of therapy, drop out and side effects of therapy. Overall, reporting of harm was inconsistent across studies and the quality of data collection and reporting varied. MBIs for psychosis appear to be safe and may reduce the risk of hospitalisation and use of crisis services. However, the absence of thorough reporting on harm precludes a balanced analysis of benefits versus harms. Future research into the effectiveness of MBIs should consistently operationalise, monitor and report data on harm.