OpenAI reports improved mental-health crisis safeguards
OpenAI reported new GPT-5 crisis-response safeguards, including safe-completion training and more than 25 percent fewer non-ideal emergency responses than GPT-4o in its evaluation.
AWAY FROM DOOM26confidence 76/100
Why it moved the index
Measured reductions in unsafe crisis responses strengthen a deployed safeguard for a mass-use frontier model, though the evidence is developer-run and domain-specific.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
AUDIT TRAIL
Assessment history
- R1Away 26 · confidence 76
New August 2025 deployed safety evidence tied to the exact durable GPT-5 profile; no durable story collision.
12 Aug 2026