Standard transformer attention treats behaviorally distinct token roles — constraints, instructions, persona, episodic context, user content — as undifferentiated competitors in one softmax budget. We argue this design is the structural root cause of five alignment failure modes: compliance collapse, sycophancy, prompt injection, persona drift, and instruction hierarchy violations. We propose Governed Multi-Stream Attention (GMSA) — typed streams with independent softmax denominators per role — as the unified fix. A GMSA K=3 prototype (Qwen2.5-32B, trained on constraint-persistence data only) achieves a mean of 0.79 across five distinct alignment benchmarks for which no targeted training was conducted, versus 0.26 baseline and 0.44 for matched LoRA on identical data. The architectural gain over LoRA averages +0.35 (range +0.16 to +0.48 across the five tasks) and is positive on every task and across three families and six model sizes (Qwen2.5 7B–72B, Llama-3.1 8B–70B, Mistral-NeMo 12B). Capability metrics are unchanged (MMLU, HumanEval, MT-Bench). The experimental anchor is Constraint Routing Failure (CRF): at Ku=8 constraints and depth 48, compliance falls to C-PP=0.07–0.22 while factual recall holds at 0.54–0.91 across six frontier models. The mechanism is architectural: per-constraint attention mass dilutes sublinearly, mi = Θ(Kγ−1 u ), γˆ=0.39 (R2=0.98), replicated across four architecture families (γˆ ∈ 0.341, 0.390). A Pinsker-based bound connects mass to enforcement gain: gi ≤ p Cθ/2·mi (Theorem 2). Causal attention surgery quantifies routing failure as ∼78% of the compliance cliff; diffuse encoding loss is a minor co-contributor (21.7%). Inference-time: a three-system cognitive architecture (Brain v1 working memory; Brain v2 episodic memory; Brain v3 executive attention) achieves mean C-PP = 0.678 across six models (+0.51 over baseline), generalizing to naturalistic sessions (+0.23 on WildChat and LMSYSChat-1M). Suppression constraints are an identified open problem (0.41 0.34, 0.48); GMSA reduces but does not fully close this gap. GMSA is proved to be the unique attention variant satisfying Behavioral Mass Invariance (∂mbeh/∂|Sctx| = 0), empirically verified at 0.90–1.00× across contexts of 200–11,400 tokens. GMSA is being integrated into Neptyn (our trillionparameter sparse MoE production model); early internal evaluations are encouraging, and a full Neptyn report is forthcoming.
alsufi et al. (Fri,) studied this question.