Direct Interaction

Definition

Direct interaction describes an exposure pathway where harm occurs because a person directly uses, engages with, or communicates with an AI system, and that engagement itself causes or transmits the harm. The victim is an active participant in the interaction — they input data, receive output, follow recommendations, or otherwise engage with the system’s interface.

This is the most common exposure pathway for consumer-facing AI systems. Examples include chatbots providing dangerous advice, AI assistants generating harmful content in response to user queries, or AI-driven interfaces that manipulate user behavior through design patterns.

How This Pathway Operates

Direct interaction harm typically follows a sequence:

1. The victim initiates or participates in an interaction with an AI system
2. The system produces output, recommendations, or behavioral nudges during that interaction
3. The victim acts on, is influenced by, or is otherwise harmed through the system’s response

The distinguishing feature is that the victim has agency in the interaction — they chose to use the system (or were required to). This contrasts with algorithmic decision impact, where victims may never interact with the system that affected them, and adversarial targeting, where a threat actor weaponizes AI against specific victims.

Distinguishing From Other Pathways

• vs. Algorithmic Decision Impact: Direct interaction requires the victim to engage with the system. Algorithmic decision impact affects people through backend scoring, filtering, or ranking they may never see.
• vs. Adversarial Targeting: Direct interaction involves the AI system operating as designed (even if poorly). Adversarial targeting involves a threat actor weaponizing the AI against specific victims.

Incidents involving this exposure pathway are listed in the sidebar.