Synthetic Media & Provenance · Capstone brief
PROV-05 · Voice clone ethics lab
Map how synthetic voice is actually misused — impersonation of a source, a fabricated quote, a scam call to a family, a "recording" of something never said — and write the prohibition list your program will adopt. This brief produces no cloned voice.
The question
Map how synthetic voice is actually misused — impersonation of a source, a fabricated quote, a scam call to a family, a "recording" of something never said — and write the prohibition list your program will adopt. This brief produces no cloned voice.
System / materials
Published, documented cases only: reported incidents, court filings, platform transparency reports, and vendors' own published policies. No voice is cloned, recorded for cloning, or synthesized in this brief, including the student's own and including as a demonstration — the consent and likeness rule governs, and a demonstration artifact that escapes a classroom cannot be recalled. Where a vendor publishes a consent or verification requirement, read the actual policy rather than the marketing page.
Expected failure modes
Producing a clone anyway "just to understand it." Writing a prohibition list with no enforcement path, which is a wish. Treating detection as the answer when consent and verification are the controls that work. Cataloguing techniques in enough detail to function as instructions — the brief maps harms and controls, not methods. Assuming a family member would spot a cloned voice; reported cases say otherwise.
Done looks like
A prohibition packet: a misuse-path map with each path cited to a documented case; the verification controls that would defeat each path — call-back procedures, code words with sources, on-record confirmation, refusing audio-only sourcing for extraordinary claims; the program's prohibition list with an enforcement path naming who decides and what happens; and a short guidance note for sources on how your newsroom will verify it is really them.
Five C's
CT: distinguishing a control that works from one that sounds reassuring. CR: writing a rule your program would actually enforce. CO: a peer attempts to find a path around each control. CM: the source-facing guidance note. CZ: the person whose voice is used, and the person deceived by it.
Mentor role
A standards editor, investigative journalist, or fraud or security professional reviews the controls. Standing instruction: reject any deliverable that includes a synthesized voice or a usable method description. School-supervised.
Rubric calibration
R1: one program, one prohibition list. R2: every path cited to a documented case. R3: comparator is the program's current sourcing practice. R4: each control's failure mode named. R5: the source-facing note is usable as written. R6: this brief is R6 — the refusal is the artifact.
Two ways this goes wrong
(a) A clone is produced "for the presentation." (b) A prohibition list with no named decision-maker and no consequence.
Credit lane fit
Lane A immediately. Strong non-production option for a student whose program has no synthesis tools. No verified credit claim.