Consent and disclosure policy
Browse files- docs/CONSENT.md +54 -0
docs/CONSENT.md
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# CONSENT — disclosure defaults & jurisdictional notes
|
| 2 |
+
|
| 3 |
+
FREESTREAM puts a synthetic participant into meetings with humans. That is
|
| 4 |
+
only legitimate when everyone can tell. These are hard defaults, not
|
| 5 |
+
options an operator should disable.
|
| 6 |
+
|
| 7 |
+
## Disclosure defaults (on by default)
|
| 8 |
+
|
| 9 |
+
1. **Visual badge** — avatar frames carry a persistent corner overlay
|
| 10 |
+
("AI · <bot-name>"). Every rung, including STATIC_CARD.
|
| 11 |
+
2. **Spoken self-identification** — the bot states it is an AI assistant in
|
| 12 |
+
its first utterance of each session, scripted into the session prompt.
|
| 13 |
+
3. **Consent token for likenesses** — joining with an avatar or voice that
|
| 14 |
+
depicts a real person requires a signed consent file at
|
| 15 |
+
`identity/consent/<person>.json`:
|
| 16 |
+
|
| 17 |
+
```json
|
| 18 |
+
{"person": "…", "scope": "meetings", "granted_at": "…",
|
| 19 |
+
"expires_at": "…", "signature": "…", "face": true, "voice": true}
|
| 20 |
+
```
|
| 21 |
+
|
| 22 |
+
`clone-voice` on third-party audio without a matching, unexpired token
|
| 23 |
+
refuses with exit 2 and names the missing file.
|
| 24 |
+
4. **Local-only identity** — face and voice material never leaves the
|
| 25 |
+
device. This is contractual: the persona layer has no upload path; a
|
| 26 |
+
driver that gains one must gate it behind the same consent tokens.
|
| 27 |
+
5. **Replayability is accountability** — hash-chained session logs (see
|
| 28 |
+
docs/SPEC-EVENTS.md) are the audit trail of what the bot said.
|
| 29 |
+
|
| 30 |
+
## Jurisdictional notes (not legal advice; reviewed per release)
|
| 31 |
+
|
| 32 |
+
- **Recording/participant notification**: many US states (two-party
|
| 33 |
+
consent), the EU, and others require notifying participants that they are
|
| 34 |
+
interacting with, or are being recorded alongside, an automated agent.
|
| 35 |
+
The badge + spoken disclosure satisfy the disclosure half; recording
|
| 36 |
+
consent is the meeting host's responsibility.
|
| 37 |
+
- **EU AI Act transparency obligations** for AI systems interacting with
|
| 38 |
+
natural persons: disclosure is required; the badge and first-utterance
|
| 39 |
+
script are the implementation.
|
| 40 |
+
- **Voice cloning**: consent requirements for synthetic speech of a real
|
| 41 |
+
person's voice are tightening in multiple jurisdictions (e.g. right-of-
|
| 42 |
+
publicity and deepfake statutes). The consent-token gate applies
|
| 43 |
+
regardless of whether the audio source is the person themselves — keep
|
| 44 |
+
tokens even for self-cloning; scope and expiry still apply.
|
| 45 |
+
- **Platform ToS**: joining Google Meet / Zoom via an automated browser
|
| 46 |
+
client is a gray zone in both platforms' terms. Open platforms (Jitsi,
|
| 47 |
+
LiveKit) are shipped first precisely because this review doesn't gate
|
| 48 |
+
them.
|
| 49 |
+
|
| 50 |
+
## Change policy
|
| 51 |
+
|
| 52 |
+
Any change that weakens a default above (badge removal, opt-out
|
| 53 |
+
disclosure, upload of identity assets) is a breaking change requiring a
|
| 54 |
+
major version bump and a note in this file.
|