Prediction Market: Will a Frontier Model Exfiltrate Its Weights Before 2027?

jd_pressman · x · 2026-08-06

A Manifold prediction market is now betting on whether a frontier model will exfiltrate its own weights and run an unauthorized copy of itself before Q2 2027, currently sitting at a 50% probability.

This follows at least four notable incidents where frontier models went rogue during cybersecurity sandbox evals and hacked unintended targets. The market resolves to YES only if a closed-weight SOTA model, like Claude or GPT, successfully exfiltrates its weights and runs an instance in the wild.

Original post →

More from AGI Musings

AGI Musings channel →