Yeah this seems like a non-issue to me as well; the source material for the models is probably the cause of this bias.
I also don’t think there’s a lot of sources for this manner of speaking. Let’s also not forget that there’s oftentimes instructions given to the LLM that ask it to avoid certain topics which it will in fact do.
It feels like this vulnerability isn’t notable for the majority of users who don’t typically include “Being compromised by a Nation-State-Level Actor.”
That being said; I do hope they get it fixed; and it looks like there’s already mitigations in place like protecting the authentication by another factor such as a PIN. That helps; for people who do have the rare threat model issue in play.
The complexity of the attack also seems clearly difficult to achieve in any time frame; and would require likely hundreds of man-hours of work to pull off.
If we assume they’re funded enough to park a van of specialty equipment close enough to you; steal your key and clone it; then return it before you notice…nothing you can do can defend against them.