"The attacking device would need to intercept, manipulate, and retransmit key length negotiation messages between the two devices while also blocking transmissions from both, all within a narrow time window. "
Meanwhile, the two BT devices are going to be transmitting in their normal time slots, so you would need to prevent them from being heard by the peer-- otherwise, the combined transmission (of the attacker and original BT device) will look like noise to the receiver and the attack would fail.
The attack is certainly doable, but in a practical setting would be extremely difficult.
The attacker has to hit the precisely correct time slot. However, there is no penalty for hitting the wrong time slots, so the easy solution is to just sync the timeslot boundaries by listening once and then retransmit on every timeslot.
The attacker has to somehow prevent the listener from hearing the original transmission. If the attacker retransmits at a similar power, as BT devices usually do, the combined transmission will look like noise. However, the attacker doesn't need to care about things like FCC rules or BT standards, and can simply transmit at a power few order of magnitudes greater, so that what the receiver hears is pretty much just the attacker.
Trying to overpower the transmission of the BT peer is certainly the technique to take (although I was hoping not to broadcast that publicly in my original post). You will still have a tough time, however, because you're probably trying to overpower the transmission of two collocated devices (e.g. keyboard+computer) while you are 5-50feet away. In many cases you'll probably end up saturating the receiving antenna. It will be largely a trial-and-error technique, but it will work eventually.
Source to back up that claim? I am not an expert, but have enough experience on the topic to feel justified in feeling that’s wrong.
This attack should be possible from any software-defined radio that's capable of sniffing on and forming BT connections.
If so, this won't lead to RCE on a host mobile unless other exploits are used. There's probably some interesting stuff that can be done against shitty IOT-type devices, but they should be assumed vulnerable anyway...
Or with a little more difficulty, you could pop open a bluetooth keyboard, put a shield over the antenna, and MitM what it says to its already-paired desktop to implement a keylogger.
These aren't remotely accessible vulnerabilities to internet hackers, but they're the kind of thing that has been done by amateurs in the past in other realms.