Correct Safety section to claim only what the product does #160

Open

will.anderson wants to merge 1 commits from fix/safety-claims-match-shipped-behavior into main

Author	SHA1	Message	Date
will.anderson	bb419dc98c	Correct safety section to claim only what the product does The Safety section described crisis capabilities the product does not have. Grounded against products/neuron/safety.el and the desktop safety flow, the shipped behavior is: on-device keyword detection that surfaces crisis resources (988, hotlines) to the user, augments the assistant's reply with warmth and resources, uses abuse-aware guidance that never points a victim at their own contacts, stores a Hard Bell contact locally (never uploaded), and logs bell events on-device only. Nothing is transmitted; no one is contacted. Corrected the false claims: - "content routes to emergency services and crisis lines" -> it surfaces crisis resources to the user (card: Emergency routing -> Crisis resources) - "emergency services or relevant authorities may be contacted ... cannot opt out" -> no auto-reporting exists; card now states on-device, never transmitted (Mandatory reporting -> On your device) - "that contact is reached directly" / "the contact is made by your local instance" -> the contact is stored locally; there is no dispatch mechanism - "I'm also establishing a Neuron crisis line - free, 24/7, staffed" -> no such service exists; removed - reframed the "routing" copy that justified the above to the real abuse-aware guidance the HHS stat actually supports Source only. Site build/deploy is held; dist and index.html are regenerated at build time and are unchanged here.	2026-07-21 09:50:12 -05:00

Author

SHA1

Message

Date

will.anderson

bb419dc98c

Correct safety section to claim only what the product does

The Safety section described crisis capabilities the product does not have.
Grounded against products/neuron/safety.el and the desktop safety flow, the
shipped behavior is: on-device keyword detection that surfaces crisis
resources (988, hotlines) to the user, augments the assistant's reply with
warmth and resources, uses abuse-aware guidance that never points a victim at
their own contacts, stores a Hard Bell contact locally (never uploaded), and
logs bell events on-device only. Nothing is transmitted; no one is contacted.

Corrected the false claims:
- "content routes to emergency services and crisis lines" -> it surfaces
  crisis resources to the user (card: Emergency routing -> Crisis resources)
- "emergency services or relevant authorities may be contacted ... cannot opt
  out" -> no auto-reporting exists; card now states on-device, never
  transmitted (Mandatory reporting -> On your device)
- "that contact is reached directly" / "the contact is made by your local
  instance" -> the contact is stored locally; there is no dispatch mechanism
- "I'm also establishing a Neuron crisis line - free, 24/7, staffed" -> no
  such service exists; removed
- reframed the "routing" copy that justified the above to the real
  abuse-aware guidance the HHS stat actually supports

Source only. Site build/deploy is held; dist and index.html are regenerated
at build time and are unchanged here.

2026-07-21 09:50:12 -05:00

Correct Safety section to claim only what the product does #160

1 Commits