Gait recognition systems can start with beneficial intentions, like monitoring fall risk in elderly patients. However, without strict boundaries, these systems can lead to function creep, where data is used for unintended purposes, such as insurers flagging applicants as high-risk without consent. This trajectory is not hypothetical but a foreseeable risk of deploying powerful recognition systems without hard limits. Last lecture established that privacy by design is architecture, not a disclaimer. Now the key idea is what happens when design fails, or when design is deliberately bypassed. Gait recognition is biometric identification the moment a system technically processes walking behavior to establish or confirm a person's identity. Under Article 9 of GDPR, that data is a special category. Processing is generally prohibited unless a specific legal exception applies. The EU AI Act goes further: it explicitly names gait and posture among characteristics usable for biometric identification. That is not a technicality. It is a legal boundary with real enforcement weight. Consent becomes complex in public spaces. Recording in a transit hub does not equate to informed consent for biometric analysis. With thousands passing through, opting out is impractical. Gait can be captured without knowledge or participation, making its powerful non-cooperative quality ethically challenging when deployed without constraints. A recognition score is not proof of identity. False positives wrongly associate one person with another. False negatives fail to recognize the correct person. In policing or border control, a false positive can mean wrongful detention. In healthcare, a false negative can mean a missed intervention. System evaluation must report both rates separately, not blend them into a single accuracy number. And here is the number that should stop anyone from over-trusting a benchmark: one experiment found rank-one recognition drop from 78 percent to 3 percent when shoe type and a six-month time gap were introduced. Same system. Radically different real-world performance. Gait data exposes more than identity, Jordan. It can reveal age, sex, ethnicity, and physical or neurological health. In one mobile mixed-reality setting, gait profiling reached 78 percent accuracy for a sensitive-attribute inference task on unprotected video. That means a system built for access control can simultaneously profile health status. The EU AI Act explicitly prohibits biometric categorization systems designed to infer political opinions, religious beliefs, or sexual orientation from biometric data. The prohibition exists because the inference is technically possible. Possibility without prohibition becomes practice. Biometric templates are not passwords, Jordan. A leaked gait template creates lasting risk because the characteristic it encodes cannot simply be replaced. You cannot choose a new walk the way you choose a new PIN. Beyond individual harm, large-scale gait recognition changes the practical meaning of anonymity in public space. Remote biometric monitoring can chill freedom of expression and assembly. People modify behavior when they believe they are being watched. That behavioral change is itself a harm, even when no individual is ever misidentified. A system's high laboratory accuracy does not guarantee ethical deployment. Controlled testing often overstates real-world performance. Human reviewers may over-trust outputs without uncertainty information and authority to reject results. Responsible governance requires impact assessments, independent audits, public notice, and procurement limits to prevent function creep. Privacy protection must span the full lifecycle: collection, training, deployment, and retirement. The takeaway is this. Consent and transparency are the guardrails that prevent biometric movement analysis from becoming a tool of oppression. Without them, the same technology that monitors a patient's recovery can surveil a protest, profile a job applicant, or flag a traveler, all from the same camera, reading the same walk.