Not All Concepts Are Causal: Sparse Autoencoders Reveal a Motor Layer in VLAs

Vision-language-action (VLA) policies hide which internal features merely describe a task and which produce the action. We introduce SAE-VLA, which turns a frozen VLA's activations into a concept lookup table for reading out and steering its actions. A sparse autoencoder provides the rows, a vision-language model names them, and a blinded verifier checks every name on held-out episodes. A signed intervention records whether a row moves the action its name implies, at a dose certified by a closed-loop gate. Across five policies on two backbones, where the table is built decides whether it works. In ${\pi }_{0.5}$, language features receive plausible motor names that verify half as often and barely act. The action expert instead holds a motor layer of grasp, release, hand-height, and wrist-tilt features. On the ${\pi }_{0.5}$ breakfast policy, verified names identify the moving arm on $98.6\%$ of held-out single-arm frames and contain the current subtask on $97\%$, by lookup alone. A name indicates where to steer but not whether, so the table's causal column defines what is steerable. On the ${\pi }_{0.5}$ LIBERO policy, a gated verified row makes the arm descend earlier with success unchanged. Gated handles replicate with a second dictionary and on the OpenVLA-OFT LIBERO policy. Steering by language lets the frozen ${\pi }_{0.5}$ breakfast policy recover bread placements beyond its demonstrations, raising task success from $17.5\%$ to $87.5\%$ without new data or retraining. Not every named concept is causal; the table records which ones are.