What happened
Forbes reports that Sanctuary AI is expanding beyond its Phoenix humanoid robot by offering its Physical AI system for use with existing industrial robots. The system was tested on a difficult wire-plugging task for an unnamed global Tier 1 automotive supplier, where robotic hands inserted flexible wired plugs into moving targets on a live conveyor.
Forbes reports that Sanctuary AI recently achieved a 99.5%-plus success rate and a 2.54-second cycle time on a wire-plugging task for a global Tier 1 automotive supplier. The job involved inserting flexible wired plugs into moving automotive targets on a live conveyor. According to the report, the task had not previously been automatable with traditional methods. The company used two robotic hands rather than its full Phoenix humanoid robot. The customer was not named in the source provided.
Sanctuary AI CEO Daniel Friedmann told Forbes that the reported result came from a 40-minute test involving 313 plug-insertion trials. He said the success rate was calculated by comparing successful and unsuccessful insertions. Friedmann also told Forbes that the 2.54-second cycle time was measured against the customer’s existing line performance, rather than against a named human benchmark or a competing automation system.
The company is presenting the work as an expanded, hardware-agnostic approach to Physical AI. Forbes reports that Sanctuary can run the system on currently available industrial robots, using both off-the-shelf and custom end effectors. The company’s stated goal is to deploy the technology at customer sites within a few weeks, although Friedmann acknowledged that actual timelines vary with task complexity. He described the system as suitable for contact-rich and dexterity-intensive work in manufacturing, logistics, and other labor-constrained industries.
Forbes frames the move as a possible pivot, or an addition, to Sanctuary AI’s humanoid strategy. The company still maintains Phoenix, described in the report as a humanoid platform with a torso, arms, head, and wheeled base. Friedmann told Forbes that the industrial-robot deployment is not a withdrawal from humanoids, but a way to apply the company’s AI on existing commercial platforms while building toward future intelligent robotic systems. The source does not provide a public product price, a named production customer, or evidence that the system is already operating in a factory after the test.
Why it matters
The approach could lower the hardware barrier to industrial AI by allowing companies to use existing robots rather than waiting for humanoid platforms to become widely commercialized. However, the reported result comes from a limited proof of concept, and the source does not independently verify Sanctuary AI’s performance or business claims.
The immediate significance is architectural and commercial. Humanoid robots combine a general-purpose body with an AI control system, but they also require new hardware, safety validation, maintenance processes, and capital investment. Forbes reports that Sanctuary is trying to sell the control layer and robotic dexterity separately, allowing customers to retain industrial equipment they already operate. If that model works, adoption could depend less on the arrival of mass-market humanoids.
The reported task also illustrates where industrial AI may face a meaningful technical hurdle. Plugging flexible components into moving targets requires perception, timing, force control, and hand coordination. Forbes says the work was previously out of reach for traditional automation, while Sanctuary attributes the result to its Physical AI system and robotic hands. The test therefore points to a potentially useful application area, but it does not establish that the system can solve a broad class of factory problems.
The result should nevertheless be read with careful limits. The reported 99.5% success rate covers 313 trials during a 40-minute test, not a long-term production record. The source gives no independent evaluation, detailed failure breakdown, safety analysis, maintenance data, or comparison with the best available specialized automation. It also does not identify the supplier or disclose whether the test was conducted under conditions representative of continuous factory operation. AI Understanding has not independently confirmed the figures.
For workers and employers, the implications are practical but unresolved. Sanctuary told Forbes that its focus is maintaining throughput at lower cost amid labor shortages and rising operating expenses. That could mean assistance with difficult or repetitive tasks, but it could also alter staffing needs or shift workers toward supervision and exception handling. The source provides no measured return on investment, employment effect, injury data, or evidence of deployment at scale, so those consequences remain unknown rather than demonstrated.
What to watch next
The key question is whether Sanctuary AI can convert the reported demonstration into repeatable production deployments across multiple factories and tasks. Watch for independent validation, named customers, deployment results, safety information, operating costs, and evidence that performance extends beyond the tested wire-plugging application.
The first test is whether the proof of concept becomes a production installation. Sanctuary’s CEO told Forbes that the system is not yet in production. Future evidence should show whether the same system can run for substantially longer periods, maintain its success rate under normal operating conditions, and recover safely from jams, misaligned components, damaged parts, or changing conveyor speeds. Those details would matter more than a short controlled demonstration.
The second issue is generalization. Sanctuary says its Physical AI is designed for dexterity-intensive tasks across manufacturing, logistics, and other industries, but the source documents only the wire-plugging test. Watch for additional tasks, different parts, other robot arms, and independently measured results. It will also be important to distinguish a reusable software system from a deployment that requires extensive custom engineering for each customer.
Economics will determine whether the hardware-agnostic model is commercially meaningful. The company did not provide typical customer ROI, saying it was too early to share specifics. Future reporting should establish licensing or service costs, integration time, required sensors and end effectors, maintenance needs, throughput compared with existing methods, and the conditions under which customers recover their investment. A stated goal of deployment within weeks is not evidence that every customer can meet that timeline.
Finally, watch how Sanctuary balances its commercial software strategy with its humanoid development. The company says the two efforts are symbiotic: existing robots provide hardware and training data, while Physical AI makes that hardware more capable. That relationship could create a path toward broader robotic systems, but it could also reveal tradeoffs between task-specific industrial deployments and general-purpose humanoids. The source does not establish which path will dominate or whether the reported performance will transfer to Phoenix.