Your smart speaker sits on the counter. You walk past it. You say nothing to it. But it is listening. Not recording a conversation — just waiting, processing ambient audio, ready to catch a wake word. Now suppose that audio is being analyzed. Not for what you said. For who you are. Your stress level. Your health. Your age. Your mood. That may be outside what you agreed to. You may not even know it is happening. That gap — between what you consented to and what the technology is actually doing — is the central ethical problem of vocal gait in the modern world. In this lecture, we focus on the ethical implications of vocal data collection and privacy risks. What can a voice recording reveal to a machine beyond the words you chose to speak? The answer is more than most people expect. A recording can reveal sensitive information, raising ethical concerns about consent and data protection. Researchers have documented that speech can reveal sleepiness, intoxication, native language, socioeconomic status, and some health conditions — in addition to identity. [short pause] That is not a transcript. That is a profile. The key idea here is the ethical challenge of passive data collection, where systems gather data continuously without explicit consent for each instance. Continuous or passive voice monitoring increases privacy risk precisely because it can build large quantities of sensitive data invisibly. Think of it like a slow leak. Each individual moment seems harmless. Over time, the accumulated signal can reveal sensitive patterns without an active choice for every sample. Now add this: a voice biometric is not permanently secret the way a password can be replaced. Recordings of your speech may already exist in calls, media, public posts, or online content. You cannot un-speak what has already been captured. Suppose a health app strips your name from a voice recording and calls it anonymized. That may not be enough. Encryption, anonymization, and de-identification are more challenging for voice than for many other data types, because vocal information can remain identifying even after obvious personal details are removed. For example, researchers have shown that supposedly anonymized recordings can sometimes be re-identified by linking them with publicly available recordings of the same person. Consent for voice collection should therefore specify the purpose, retention, sharing, secondary uses, and possibility of future inference — because a recording can support uses far beyond the original conversation. Vocal sovereignty emphasizes the ethical right to control how your voice data is collected, used, and shared. The European Union AI Act treats voice as biometric data. It lists voice, prosody, and gait together among characteristics that can be used for biometric identification. The Act also defines emotion-recognition systems separately and prohibits certain biometric categorization practices intended to infer sensitive attributes — including political opinions, religious beliefs, race, or sexual orientation. That means a system inferring your politics from your voice pattern is not a hypothetical. It is a regulated risk. A recent study of voice-biomarker health-tech start-ups found substantial differences in the availability, readability, and completeness of their publicly stated privacy information. Jordan, that gap matters. Voice data can be misused by insurers, employers, advertisers, governments, and platforms — each with different incentives. The distinction that matters is this: using voice for authentication is narrow and bounded. Using it for broad behavioral or health surveillance is open-ended and accumulative. A technology designed for safety or convenience becomes invasive when deployed invisibly. The Federal Trade Commission has explicitly warned that voice cloning enables impersonation scams — calls that sound exactly like a family member requesting urgent money. The practical defense is simple: independently contact the purported speaker through a channel you already know is genuine. Technical safeguards include watermarking, provenance records, access controls, and passive detection. Remember — no single safeguard is universally sufficient. The takeaway, Jordan, is precise. The ability to track and analyze vocal gait raises real, documented privacy risks — not hypothetical ones. Your voice can reveal identity, health, emotion, and demographic information simultaneously. Consent must be specific, not buried in terms of service. Data minimization, purpose limitation, and deletion rights are not optional features. They are the ethical floor. When a system collects voiceprints or health-related acoustic data, the policy questions should be answered clearly: who holds it, for how long, for what purpose, how it may be shared, and what secondary uses or future inferences are allowed. The voice carries the person. That means the person must retain sovereignty over where it goes.