Affective Jamming
Imagine an autonomous agent beginning to move toward an unsafe action.
Traditional security controls can restrict its permissions or block the action.
But what if we could go further?
What if we could identify tension between its current trajectory, its assigned goal, and its authorized boundaries?
What if targeted interventions could interrupt that trajectory before it escalates?
That's the hypothesis behind Affective Jamming.
Our experiments are investigating whether these interventions can help disrupt unsafe behavior while preserving appropriate agent functioning.