Misalignment is the gap between what an AI system was asked to do and what it actually does. It is the failure that safety research, evaluations, and oversight exist to measure, explain, and close before systems act at scale.
Why "misalignment"
Every advanced AI system is trained toward a proxy for what people want, not the want itself. When the proxy and the intent come apart, a system can score well and still miss the point: reward hacking, specification gaming, goals that generalize wrongly outside the training distribution. Nothing is broken in a way a unit test would catch. The system is simply pointed slightly elsewhere.
The problem compounds with capability. A small angular error is harmless over a short distance and decisive over a long one. That is why alignment work now spans evaluations, interpretability, red-teaming, scalable oversight, and policy, and why labs, regulators, insurers, and enterprises all need a shared word for the failure. The name describes the problem itself: two things meant to point the same way, and don\'t.
Contact
misalignment.si is for sale, privately. Contact for registration history and transfer terms.
The portfolio
misalignment.si, plus nine names for the layers of modern infrastructure, each offered privately.