If you enable the framework but never test positive (and thus never publish any of your keys), it's no different than if you had just kept Bluetooth on all the time.
If you enable the framework, later test positive, and choose to publish your diagnosis keys, each key can be used to link all your rolling identifiers together for the corresponding time period (nominally 24 hours). Contrast this with a randomizing Bluetooth implementation, which never intentionally reveals anything that would allow the different MAC addresses to be linked.
Of course, Bluetooth MAC address randomization itself is trivial to defeat for a reasonably capable and motivated adversary. If they can plant a bunch of radios for the purpose of tracking you, why can't they also use cameras?
Actually that's not true for the situation I described.
The bad actor would be able to connect any of your broadcast identifiers they observed back to each other via the diagnosis key that you published. Assuming they have a number of nodes monitoring Bluetooth traffic over a broad area that you passed through, they will be able to reconstruct the path you traveled over time.
For a naive implementation, the resolution of this reconstruction would depend on the spacing of the nodes. For a more advanced implementation, other data could be integrated to drastically improve it. Remember, your Bluetooth device is a broadcasting radio at the end of the day.
As to the likelihood of such things becoming commonplace worldwide, do bear in mind that many devices now periodically randomize their Bluetooth MAC addresses due to real world examples of tracking. Thankfully in this case it would only be possible to compromise the privacy of those who tested positive, and only within a singe 24 hour period (ie the daily tracing key rotation time frame) at that.
I'm much more concerned about reducing COVID-19 to save millions of lives.
Not quite. It takes a wiz to collect and correlate the data, yes. What happens to that data after that? For it to be useful, it's going to get stored somewhere. All it takes is an uneducated clerk or bored intern with access to go snooping around the de-anonymized data to compromise anybody implicated.
And this does happen routinely.
* Facebook, Uber and Google have all had problems with plebes (and tech wizzes!) with god-tier access doing inappropriate things with sensitive data.
* Bored data entry clerks with access to the credit reporting database routinely snoop on neighbors', exes' and celebrities' credit reports in spite of federal law.
* Revenge porn is such a thing that rule 34(a) ought to be that if you produce nudes, your confidant or Geek Squad/iRepair technician will post them on the internet.
* Look at how often people get doxxed by employees leaking customer PII onto reddit and 4chan, then look at how fast the mob descends on people innocent of any actual wrongdoing.
* We've seen a secretary get her hands on the Pepsi formula and try to sell it to Coca-Cola.
* The people living in the geographic center of America continue to receive death threats and harassment because of a flaw in outdated MaxMind databases that attributes ungeolocatable IPs to their location.
* There are people who refuse to participate in the census because of what certain cults of personality have done with such data.
Any chain of confidentiality is only as strong as its weakest link. You presume far too much intelligence and rationality on the part of humanity. Never forget that half of Americans wanted a belligerent narcissist to be "leader of the free world," and he still has supporters despite publicly recommending anti-parasitics and Lysol douches as solutions for a global viral pandemic.
Sensitive data is not created and left to decay in an underground bunker in Yuma. Despite its practical uses, at some level it will be exposed to individuals who lack discretion and will be exploited to malevolent ends.
Not once in human history has it worked out any other way!