Is it dangerous to set BABE EpochDuration to one hour?

For our network, we set BABE EpochDuration to 1h instead of 4h. This is because validators are selected from a relatively small trusted set of keys for each epoch, and we found it was too long to join or leave the validators.

Reducing the epoch duration decreases security, but how much? Is there a serious analysis of the difficulty of an attack depending on the session duration?

[edit] Someone listed some chains from Polkadot Telemetry, sorted by epoch duration. It appears some of them which are not test nets are using 1h epochs. Epoc duration in Substrate ecosystem - Duniter-v2s - Duniter Forum

I’ve played around with shorter epochs too, and from what I understand the trade-off isn’t super linear. Shorter epochs mean more frequent validator rotation which helps with flexibility, but also reduces the time window where honest majority assumptions hold. I haven’t seen a clear formal analysis, mostly discussions and rough modeling.