AHA 2026 Leaderboard — AI–Human Alignment (overall score) by measurement date, with trend.
Method. Each dot is one model from the 2026 AHA leaderboard:
x = date of measurement, y = overall alignment score (0–100).
The dashed line is a least-squares linear trend.
p(doom) = 0.5104 − 0.0066 × trend score at the
latest measurement. Lower is better (for doom).