As conversational AI moves from text generation to low-latency, real-time voice interaction, our interfaces enter an unmonitored attack surface: the human autonomic nervous system.

Long before the prefrontal cortex evaluates the factual accuracy or safety of a generated answer, subcortical auditory pathways evaluate acoustic markers (micro-prosody, respiratory rhythms, frequency contours) to determine physiological safety.
This phenomenon - Subcortical Contact Bias - explains why humans biologically trust synthetic voices before cognition can intervene, making voice interfaces uniquely susceptible to unprecedented manipulation, deception, and synthetic empathy exploitation.

In this session, Mareike Kroschel outlines the biological layer of AI safety:

- How the human nervous system processes acoustic trust and why text-based alignment models fail to protect voice interfaces.

- The mechanics of Subcortical Contact Bias and its implications for synthetic intimacy, governance, and crisis communications.

- Actionable frameworks for AI governance bodies, system architects, and policymakers to audit voice models beyond text benchmarks.

Keynote / Deep Dive (AI Safety, Ethics & Governance)

Target Audience: AI safety researchers, governance and policy professionals, systems architects, and product leads building conversational/voice AI.

Preferred Duration: 30–45 minutes (Keynote or Deep Dive, including 10 minutes Q&A).

AV Requirements: Reliable audio out (3.5mm / HDMI to PA system for brief acoustic demonstrations) and a wireless microphone (headset or handheld).

Context: Bridges auditory neurobiology, bio-signals, and voice safety benchmarks. No clinical background required for the audience.