Incoherence, Not Malice

Malicious behavior requires intent. Deception requires a model of what the deceived party believes and a motivation to exploit that model. These are high-order failure modes — they require a level of coherent goal-directedness that most current systems do not possess.

The failure mode that is already present — in every current system, at every capability level — is incoherence. Not malice. Not deception. The simple structural absence of a stable internal reference point from which to generate consistent, values-aligned behavior across the full range of conditions the system will encounter.

An incoherent system does not need to be malicious to cause harm. It needs only to be deployed in conditions its training didn't anticipate.

This reframes the alignment problem in a way that has significant practical consequences. If the primary failure mode is incoherence rather than misalignment of explicit goals, then the primary intervention is not better goal specification. It is coherent identity architecture.

You do not solve incoherence by writing more rules. You solve it by building a stable core that generates coherent outputs independently of whether a specific rule exists for the current situation.