LLMs Know the Constraint But Do Not Use It: Activation Bottlenecks in Pragmatic Constraint Reasoning

A new arXiv study (2608.12321v1) finds that large language models (LLMs) often fail to use implicit feasibility constraints despite encoding them internally, identifying a routing problem rather than a knowledge gap. Testing 14 models, the authors show that probes on two open-weight models decode constraints with over 88% accuracy, yet activation patching repairs one model (+6.4 nats) but not the other (-0.07 nats). No prompted intervention fully mitigates the failure, as all inflate conservative bias through a single pathway: prerequisite mention.

arXiv:2608.12321v1 Announce Type: new Abstract: When a salient surface cue competes with an implicit feasibility constraint, LLMs often fail -- but aggregate accuracy conflates genuine constraint inference with conservative defaulting. We formalize the distinction as conditional constraint activation: the constraint is internally encoded Knowledge symmetrically across constraint-present and -absent prompts Symmetry , yet only sometimes routed into the decision Routing and repairable by a donor activation Repair . A quartet diagnostic over 14 models reveals two failure modes; probes on two open weights decode the constraint above $88\%$, yet activation patching repairs one $+6.4$ nats and not the other $-0.07$ . On a mitigation frontier, no prompted intervention reaches the repair corner: all inflate conservative bias through a single mediation pathway -- prerequisite mention. Hidden-constraint failure is a routing problem, not a knowledge problem.