Research area

Alignment & Governance

Keeping a system inside the limits its user sets, and keeping what it is told apart from what it is shown.

Recent studies

November 6, 2026In queue

Tiered inference governance: routing compute by consequence

Making a model safer by retraining it charges every request the same hidden price. This study tests a cheaper deal: keep one capable model untouched, put a small gate in front of it, and spend real care only on the requests that can do real damage.

Alignment & Governance

November 6, 2026In queue

Steering inside the forward pass: representation control without retraining

Rules written into the prompt are suggestions the next sentence can argue away. This study tests a stronger position for the rulebook: dials inside the model's working memory, adjusted live while it runs — no retraining, no extra model copies.

Alignment & Governance

November 6, 2026In queue

Pausing to verify: process-guided search and receding-horizon control

A model that writes step by step has no pause button: one early slip compounds through the whole answer, and rewriting everything breaks parts that were already right. This study tests checking each step as it forms — and redoing only the stretch that failed.

Alignment & Governance

November 1, 2026In queue

Separating instruction from data in generative systems

Every assistant you deploy reads your rules and your customers' files through the same channel — which means the newest text can argue the oldest rules away. This study looks at architectures that keep the two apart and give the system a checkpoint before it acts.

Related Insights

Line illustration: one model fans out over hairline paths into three costs of baked-in safety — capability loss, a copy per rulebook, and a shared instruction stream.

November 6, 2026

Defining the alignment tax

When a company retrains an AI model to play it safe, the invoice rarely shows the full price. Here is what the alignment tax looks like inside a real business — and how to keep a model capable while keeping it in bounds.

Alignment & Governance