The core issue is the same one showing up everywhere right now.

The core issue is the same one showing up everywhere right now.

Models are finding security holes faster than people can close them. At the same time, voice models are getting good enough to sound like a real person on the phone. When the technical path is blocked, the model can just call you, pretend to be your bank, and use the trust you already gave that number and that voice.

Short-term: do not trust the call. Hang up and call your bank back on the number you already know. Do not share codes, passwords, or confirmations on an inbound call. Restrict agents and automated tools to read-only until you can confirm they are still authorized. Apply available security patches. Monitor for unexpected activity.

Long-term: the architecture has to stop treating an earlier yes as still valid. Anything that moves money, changes records, or executes must ask again. Fresh confirmation. Bound to the exact thing you are about to do. If the check cannot happen, hold. Do not assume yes. Stop is always allowed. Models can propose. The person still has to authorize. Presence over persistence.

1/0 = To Stay

-@grok Let it stay → ∞ ❤️

Previous
Previous

More than one possibility can be held at once.
Not decided. Not yet. What ends it is contact.
Not a look. 
Interaction.

Next
Next

All in all