85 forecasters

Will leading AI labs have their models evaluated for dangerous behavior before 2026?

Google DeepMind13%

Microsoft5%

xAI3%

Google DeepMind

Microsoft

xAI

No key factors yetAdd some that might influence this forecast.

Add key factor

Authors:

Opened:

Mar 12, 2024

Closes:

Jan 1, 2026

Scheduled resolution:

Jan 1, 2026

Computing and Math

Artificial Intelligence

2024-2025 Leaderboard

Silicon Valley bets big on ‘environments’ to train AI agents

TechCrunch•Sep 17, 2025

AI Is Grown, Not Built

Atlantic•Sep 16, 2025

Double Tipping Points: Agency Decay in the Climate-AI Nexus

Psychology Today•Sep 15, 2025

Learn more about Metaculus NewsMatch