The humanoid company that stopped waiting for humanoids — Sanctuary puts its AI on industrial arms

Sanctuary AI spent years building one of the most ambitious humanoids in the industry: Phoenix, a robot designed around hydraulic hands with 20 degrees of freedom and haptic feedback, trained through high-fidelity teleoperation. On 17 June 2026 the Vancouver company announced that the AI it built on that data is going to work somewhere else — on industrial robot arms that factories already own. The demonstration is deliberately unglamorous. On a wire-plugging task with a global Tier 1 automotive supplier, Sanctuary reports a success rate above 99.5% at a cycle time of 2.54 seconds, which it says matches the throughput of the customer's live production line. Wire plugging is exactly the kind of job that has resisted automation: parts are flexible, tolerances are tight, and the feedback is tactile rather than visual. The reasoning behind the move is stated plainly by co-founder and CTO Olivia Norton: "Physical AI adoption is gated by AI that meets both performance and cycle time requirements. That's what customers are seeking, and that's what we are delivering." In other words, the constraint is no longer legs or hands — it is software that hits an industrial cycle time. Sanctuary now calls its approach hardware-agnostic and frames humanoids as a later step rather than the entry point. The eighth generation of Phoenix, released in December 2024, already pointed this way: it was built for high-quality data capture, with an upgraded telemetry system, wider-field and higher-resolution cameras and a cheaper bill of materials, not as a product to sell. Daniel Friedmann was appointed the company's CEO in June 2026. For a sector that has spent two years measuring progress in humanoid unit counts, this is a notable dissent from a company that helped set that agenda: the money, Sanctuary is arguing, is in the model — and the model does not care what body it runs on.