Most alignment discourse still focuses on individual agents: how do we align the model, how do we prevent hallucinations, how do we enforce correct behavior? This framing is obsolete. Modern AI systems are multi-agent ecosystems. And in these ecosystems, the real alignment problem is not the agent, the model, the prompt, the rule, or the constraint. The real alignment problem is the coalition. Coalition drift — not individual drift — is the existential governance risk. Coalitions produce coherent misalignment. Coalitions don’t just drift. They drift together. And they drift together in ways that are invisible to single-agent governance, reinforced by coordination, amplified by inference, and inevitable in open ecosystems. Alignment must target the system, not the agent. The coalition is the real alignment surface. This is the central claim of this paper and the pivot point of the Coalition Drift Series.
Building similarity graph...
Analyzing shared references across papers
Loading...
Narnaiezzsshaa Truong
American Rock Mechanics Association
Building similarity graph...
Analyzing shared references across papers
Loading...
Narnaiezzsshaa Truong (Sat,) studied this question.
www.synapsesocial.com/papers/69db37ca4fe01fead37c5d0c — DOI: https://doi.org/10.5281/zenodo.19503152