San Francisco startup Goodfire released Silico, a mechanistic interpretability tool that maps neurons and pathways inside LLMs and lets developers tweak them to reduce unwanted behaviors. In testing, Silico cut hallucinations by 58% with ~90x lower cost per intervention than LLM-as-judge approaches, with no benchmark degradation. MIT Technology Review named mechanistic interpretability one of its 10 Breakthrough Technologies of 2026; Silico works on open-source models today.