"Can be accessed" is slightly exaggerated. All another phone can get from it is "I'm nearby and my code is xxxxxxxxxxx". The other phone then sends to Apple "there's a phone near location ********** and it's code is xxxxxxxxxx" without identifying itself in any way.
It's even less exploitable than that - what the other (relaying) phone sends to Apple is "there's a phone whose device id hashes to yyyyyyyyyy and that phone's encrypted location is zzzzzzzzzz" - it doesn't have to identify itself, or the location.
Apple can't decrypt zzzzzzzzzz (the relaying phone used the public key broadcast by the target phone to encrypt their shared current location), and neither Apple nor the relaying phone can un-hash yyyyyyyyyy to find out who/what that other device is. Apple just stores the yyyyyyyyyy => zzzzzzzzzz pairs in a giant database for some period of time.
Also, yyyyyyyyyy is a value that changes periodically (one guess is every hour, though it sounds like that hasn't been announced) - it changes in an unpredictable pattern based on a secret seed value known only by several Apple devices owned by one person (in order for this whole scheme to work, you need to own more than one participating device - iPhone, iPad, Mac, maybe the Watch?). So even if BadActor/Agency were to work out the device hash for a target user (say they had an agent across the room who sniffed the BlueTooth traffic), that won't tell them where said individual was an hour ago, or where they are an hour from now, when the device will have moved on to another unguessable hash. The hashing of the device id works sort of like 2FA, using a secret seed value (shared among a user's devices) and (likely) the current time (to the minute or hour, or some other granularity) and a particular device's device id, to compute yyyyyyyyyy.
Those devices (that one person owns) - and
only those devices - know that secret shared seed value, and the private key needed to decrypt locations. Apple only has a warehouse yyyyyyyyyy => zzzzzzzzzz tuples. When one of a person's devices goes missing, they use their other device to do a Find My search. That device says, "Apple, here's a list of recent yyyyyyyyyy hash values, please give me any corresponding zzzzzzzzzz (
encrypted location data) values". It gets back a list of perhaps a few dozen zzzzzzzzzz values. These are encrypted. It tries to decrypt them, all using the private key that only the owner's devices know. Now, it may only be able to successfully decrypt some of them (a given hash value could potentially have been generated by your phone today, and by someone else's phone last week). No matter, it discards any that don't decrypt properly, the rest give it a validated list of "your sibling device D was very near location Z at time T" points, and it can draw a map for the user to interpret. All without anyone but the owner's devices ever having access to the location information in any useful way.
Apple keeps a big database of yyyyyyyyyy => zzzzzzzzzz pairs, with a timestamp on each one, it uses the timestamp to expire out the key/value pairs after some period of time (otherwise the database would grow unmanageably large). If someone (say a local or foreign government agency) demands information, all Apple can give is a pile of yyyyyyyyyy => zzzzzzzzzz pairs that are of no use to anyone but the device's owner - a demanding agency can't make heads or tails of it, there's no easy way back from the encrypted/hashed tuples, unless you have one of the owner's participating devices.
From the articles I've read that describe the procedure, it sounds pretty foolproof and safe (it depends, of course, on how well Apple implements it all, but Apple seems highly motivated to get it right). All the "oh noes! Apple will track everyone's location and bad guys/governments will use this to track me" chatter is coming from people who have only read the headline (or information-free articles) and made a lot of incorrect assumptions.
