← product

Downstream Glasses — informed-use acknowledgment (DRAFT for counsel)

Status: DRAFT — counsel-gated. Nothing here ships until reviewed (same gate as JQ_SPEECH_MIRROR.md and LEGACY_CHANNEL_ELDERCARE.md). Review scope is item (4) of product task c09bd5aa (ToS/Privacy counsel review).

What prompted this

Downstream Glasses speak in the wearer's own cloned voice, in their ear. That is psychologically novel territory. There is no evidence an in-ear assistant causes psychosis — and this document deliberately avoids implying one — but for a person with a psychosis history or fragile reality-testing, an external voice that sounds like their own could plausibly be distressing or get woven into symptoms. The right response is informed use + contraindication guidance + design mitigations, not a disease waiver.

Position we're asking counsel to confirm: a specific-disease liability waiver ("not liable if you develop schizophrenia") is (a) legally weak — personal-injury negligence generally can't be waived, and naming a disease with no causal mechanism invites the inference we believe there is one; (b) alarmist copy that plants the fear it disclaims. Informed-use acknowledgment + honest contraindication language + documented design mitigations is the stronger posture. If counsel wants belt-and-suspenders, general assumption-of-risk / wellness-not-medical language belongs in the ToS (task c09bd5aa), not a standalone waiver screen.

The design mitigations (already shipped — facts counsel can rely on)

  1. The glasses never speak unprompted. Every utterance is an explicit trigger: a button press, a spoken command, or a cue the wearer configured. Calm-by-default is a hard design rule (session.ts).
  2. Everything spoken is on the record. Every nudge, receipt, cue, and soundtrack event is logged at the moment it plays and visible to the wearer at /dashboard/glasses (migration 491, PR #1288). "Did my glasses say that, or did I imagine it?" has an auditable answer — a genuine reality-testing aid, not just an audit trail.
  3. We log what the glasses said, never what the wearer said. The Speech Mirror's matched utterances never leave the device.
  4. Kill switches. The wearer can stop the voice at any time; server-side crons have one-tap admin kill switches; the voice clone itself is consent-gated (consent v2) and deletable.
  5. Own-voice is opt-in twice. The wearer records their voice clone under explicit consent, then separately sets up the glasses.

Draft setup-screen acknowledgment (the copy counsel is reacting to)

Shown ONCE at glasses setup (after voice-clone consent, before first connection). Tone rules: honest, calm, no clinical claims, no fear-planting. Checkbox, not signature. Decline = glasses work with no voice (visual/tone cues only) or not at all — counsel to advise which.

Before you put them on

Your glasses speak in your own voice. You cloned it; they use it. Most people find this motivating — it's the point. It can also feel strange, especially the first few days.

A few things we want you to know, plainly:

  • They only speak when you ask. A button press, a command, or a cue you set up yourself. Never on their own.
  • Everything they say is on the record. Your dashboard keeps a complete, replayable log of every word your glasses have ever said to you. If you're ever unsure whether you heard them — check it.
  • You can turn the voice off any time, from the glasses or the app.
  • This isn't therapy, diagnosis, or treatment. It's a productivity and encouragement tool.
  • If you have a history of psychosis, hallucinations, or dissociation — or hearing your own voice ever becomes distressing rather than helpful — stop using the voice features and talk to your clinician. The glasses work fine with the voice off.

☐ I understand what the glasses do, and I know where the log and the off switch are.

Questions for counsel

  1. Is checkbox-acknowledgment at setup the right instrument, or should this language live only in the ToS (or both)?
  2. Decline path: voice-off mode vs. no glasses at all — any liability difference?
  3. Does naming specific conditions ("psychosis, hallucinations, dissociation") help (informed) or hurt (implies causation)? Alternative: "a history of conditions that affect how you experience voices."
  4. Minors: glasses + own-voice under the ToS age threshold — allowed with guardian consent, or excluded?
  5. Interaction with the voice-clone consent (consent v2) and BIPA/biometric language already in the ToS draft — one combined consent moment or two?
  6. Coach/teammate voices (DreamPro Glasses Voice Program): does third-party voice in-ear need its own acknowledgment line, or does the recorded-clip consent cover it?

Explicitly out of scope

  • Any claim, in either direction, about causal links between voice assistants and psychotic disorders. We make none and disclaim none by name.
  • Clinical screening at onboarding (we are not qualified to screen and asking diagnostic questions creates data we shouldn't hold).
  • Changes to the Speech Mirror gate — it stays separately counsel-gated and OFF by default.
GLASSES INFORMED USE — Docs | HiveJournal