My wife has an army of bots at her command, and on Sunday, I caught her issuing them their marching orders. Don’t worry — Melissa has no plans to conquer the world, or at least not yet.
Or not that I know of, anyway.
We’ll get to that in a moment, but what got me thinking (again) about AI in recent days is the story of the Florida woman* whose digital “diary” entry landed her in jail recently.
Carli Michelle Heller — innocent until proven guilty, I think still applies here — used Anthropic’s Claude AI as a personal diary, apparently thinking that somebody else’s servers were as private as a locked journal. Gadget Review reported that “her words included a stated intention to ‘shoot up’ the Lee County Sheriff’s Office.”
That was enough to trigger Claude’s safety systems, which flagged the chat. “A human reviewer assessed the threat as credible,” GR continued, “and law enforcement showed up at her door.”
All without a warrant.
*Of course it was Florida Woman.
Needless to say, I have some seriously mixed feelings about all that, even after reading that Heller’s “subsequent message allegedly referenced obtaining a new gun.”
But as Florida Woman learned, AI chats aren’t end-to-end encrypted in the same way your more reputable messaging apps are. I mean, you aren’t using somebody else’s data center to move a secure message from your phone to your recipient’s — you’re using somebody else’s server to directly answer your questions about life, the universe, and converting an AR-15 to fully automatic.
Sure, AIs have guardrails built in to prevent users from, say, getting detailed instructions on how to weaponize a common flu virus, but those guardrails do (and should) include notifying the proper authorities when somebody queries, “How do I build a portable nuclear device in my garage?”
