If they were going to "de-identify" the data for their telemetry, then I'd need to see some rigorous mathematical proof of it, to have any trust in their promis

If they were going to "de-identify" the data for their telemetry, then I'd need to see some rigorous mathematical proof of it, to have any trust in their promises. You will eventually compromise the individual datapoints in a dataset, given enough queries. There is in fact a field of research that specifically studies just this. The PMs at 1password haven't done their homework, they're just waving their hands, and it is worrying for users.