Socio-Technical Attractor Control
Deployment environments can select for or destroy alignment properties even when a system starts in a better state.
What decision changes?
Ask what the surrounding incentives will amplify after deployment, not only what the system passed before launch.
An alignment property that survives in a lab can be destroyed by the selection environment (deployment environment): the institutions, markets, and protocols that copy, fund, integrate, audit, and replace AI systems.
Socio-technical attractor control treats deployment as a selection process measured by deployment leverage and deployment growth rate — not market cap alone, and not Demski in-optimizer “selection” or Wentworth selection theorems (homographs). Revenue, speed, regulatory pressure, user dependence, benchmark prestige, and institutional habits can all move control toward systems that are harder to correct even when preservation conditions fail.
The goal is not to find a perfect unilateral move. It is to create a basin where safer properties are easier to preserve than to abandon. That requires artifacts that travel across roles: procurement questions, audit routines, incident thresholds, release gates, and review norms.
The practical test is whether this turns into better Monday-morning questions for an organization about what it is about to deploy, not just a result that held up in a lab.
What would count as evidence?
Evidence would include points of control over deployment (selection handles), deployment-leverage measurements, audit triggers, and signs that institutions reward correction-preserving behavior.