I'm not sure that I fully track this, but:

Today, I Dathon Pwn, am publicly disclosing BlockSlop, a consensus bug in the late-upgrade path of the BIP 110 activation client

The effective treatment of an already-connected block depends on the history of the datadir, not on the block sequence and the node's current rules alone. Two nodes running identical enforcing rules can therefore disagree about the active chain

There's also a twitter post: https://x.com/DathonPwn/status/2078514978773192779