[raxxorraxor]

I don't think AI informing authorities (who would do nothing) is in any way desirable.

At some point you have to just live with marginal dangers. There is no technical solution here.

[mensetmanusman]

Wouldn’t have to be ‘authorities’, but for people with broken families, authorities who knock on the door and check for a health crisis are literally all that they have.

(We have family that would be homeless if we hadn’t taken them under our wings to house them).

[raxxorraxor]

I am sorry to hear that. But these problems would still not justify surveillance of users with a direct line to authorities. In some countries these can be quite abusive. For people that such engagement with AI is dangerous should look for other means.

[simianwords]

a nice way to convey this is: the optimal amount of such mishaps is non-zero.