TBH, I’d reject Mythos or similar reports and require full reproduction on a publicly available model before considering them valid.