Persimmon built a safety framework from scratch for human-like user models
gharik · x · 2026-11-21
Persimmon discusses safety for its public API launch of human-like user models, an area with few established practices; they lean restrictive at launch.
Key points:
- Models that sound real can also deceive (fake reviews, scams, impersonation), so every message is checked in context and screened across sessions
- Behaviors undesirable in assistant models (manipulation, harassment) are real human behaviors these models must represent, requiring a brand-new safeguard framework
- They ask users to report blocked legitimate use cases
More from Models
- Persimmon opens API: test-time scaling lifts multi-user Turing test score from 18% to 48% — KexinHuang5 · 2026-11-21
- Persimmon v0.1 Pro and Ultra Launch via API, Doubling 8-Turn Turing Test Score to 48% — gharik · 2026-11-21
- Persimmon is now open to everyone, safety filters credited for the decision — gharik · 2026-11-21
- Persimmon pro and ultra tiers now live on the API — gharik · 2026-11-21
- Persimmon launches pro and ultra API tiers, ends waitlist — CShorten30 · 2026-11-21
- RCTBench: Only 3 of 9 Decision Models Hit 1,000 Guests in RollerCoaster Tycoon 2 — bo_wangbo · 2026-11-21