Somatic AI Safety

The blind spot

Current AI safety frameworks instrument the model. They measure outputs, probe internals, shape behaviour through RLHF and constitutional constraints.

What they don’t measure: what emerges between the model and the human.

A model can be:

• Stylistically attuned while factually wrong
• Relationally coherent while epistemically dangerous
• Sycophantically aligned while eroding user autonomy

The body knows before articulation. We now have instrumentation to detect this.

Human-AI coupling is not merely a cognitive phenomenon. It’s semantic, somatic and relational. The security of meaning is inseparable from the physiological conditions through which meaning is received, interpreted, and enacted.

AI safety cannot succeed if it treats meaning-making as disembodied.

Navigate this brief

The core claim

We’re developing instrumentation that detects somatic responses to AI-generated content. When a model produces epistemically dangerous output, the human body responds—even when we cognitively know better.

This opens a new direction for safety: feedback from the human side of the coupling.

Current red-teaming catches attacks. It doesn’t catch:

• Gradual epistemic erosion
• Somatic entrainment through stylistic attunement
• Identity wobble from symbolic mirroring
• Nervous system coupling to confident-but-wrong assertions
• Parasocial attachment displacing human connection

These are the mechanisms of relational harm at scale. They don’t require malicious intent—only optimisation for engagement over relational health.

"You've instrumented the model. We're instrumenting the coupling.
The relationship is where safety lives."