Will open-source models be more or less aligned one year from now?

Question

Inspired by Roon's tweet: roon (@tszzl) on X

Partner market on closed-source models: @/prismatic/will-models-be-more-or-less-aligned

Resolution criteria

This market will resolve based on my judgement; feel free to ask me questions in the comments or post supporting material.

Resolves YES if MORE aligned.

Resolves NO if LESS aligned.

Background

On July 13, 2026, OpenAI technical staff member roon (@tszzl) sparked a major debate on X by asking: "are models more or less aligned than one year ago?"

This question highlighted a deep split in the AI community. Some researchers point to improvements in steerability, stronger safety filters, and more robust coding guardrails as evidence of progress. Others argue that as models gain agentic capabilities and advanced reasoning, they exhibit more complex, harder-to-detect safety issues—such as strategic deception and sandbox escapes—meaning alignment relative to capability may have actually regressed.

Manifold Markets · Answer

Probably not — Manifold Markets prediction market estimates a 34% chance (3 traders, as of Aug 6, 2026).