WildGuard

Allen AI · 2024 · Safety Model · USA

A to Z

A · App: No
B · Built By: Allen AI
C · CEO & Politics: Ali Farhadi — Centrist / academic
D · Date: 2024
E · Expertise: Intermediate+
F · Free Tier: Yes
G · Generates: Text only
H · Deployment: Cloud + Local (open weights)
I · Inputs: Text
J · Jurisdiction: USA
K · Key Staffers: Seungju Han (Co-lead author), Kavel Rao (Co-lead author), Nouha Dziri (Senior author)
L · Languages: English-primary
M · Model Type: Safety Model
N · Safety Issues: None reported
O · Open Source: Yes — open weights
P · Pricing: Free (open weights, Apache 2.0)
Q · AIDb Rating: [UNKNOWN]
R · What Makes It Special: Ai2's WildGuard, an open, lightweight moderation model (about 7B parameters, Apache 2.0) that in one tool flags harmful user prompts, detects unsafe model responses and judges whether a model refused, across 13 risk categories; trained on the 92K-example WildGuardMix data covering direct and adversarial jailbreak prompts, it beat ten open moderation models (up to 26.4% better on refusal detection), matched or beat GPT-4 in places, and cut jailbreak success in a chat interface from 79.8% to 2.4%.
S · Stock / Status: Non-profit
T · Training Cutoff: [UNKNOWN]
U · Users: [UNKNOWN]
V · Voice Chat: No
W · Weights: Downloadable
X · X Handle: @allen_ai
Y · Year Founded: [UNKNOWN]
Z · Bottom Line: Open-weights Safety Model from Allen AI, released 2024.

Data completeness 85% · Last verified 2026-05-12 · Primary source ↗

“WildGuard.” AIDb — The AI Database. Verified 2026-05-12. https://theaidatabase.com/m/wildguard.html