What are the risks of humanoid robots?

Physical safety: humanoids weigh 25-80 kg and operate in shared human spaces

A modern commercial humanoid robot typically weighs 25-80 kg (55-175 lb). A failure mode that ends with the robot falling or making unintended contact with a human is a serious event, substantially more so than for a fixed industrial arm with a cage around it. Specific hazards include falls during normal walking on slippery / sloped / cluttered surfaces; collisions during loco-manipulation (walking-while-carrying is the hardest combined task and the one most likely to produce unintended contact); end-effector force at the dexterous manipulation layer (insufficient or excessive force in a shared space is a hazard); and unintended behavior under perception failure (humanoids that lose tracking in low light or with degraded sensors can make decisions that look like errors to human observers). The collected incidents involving humanoids in commercial deployment are still small in number; that will change as the fielded population grows.

Teleoperation conflation: the structural verification risk

A persistent risk in evaluating humanoid claims is conflating a teleoperated demonstration with autonomous capability. When a humanoid serves a drink at a staged event, the behavior may be autonomous (robot perceived + decided + acted on its own); teleoperated (a human is providing control inputs remotely); or pre-programmed (the robot is executing a scripted routine). These are very different from a capability standpoint, and the difference is often not labeled in public demos. Tesla acknowledged that some Optimus demonstrations at the October 2024 We Robot event involved teleoperation. This is not unique to Tesla; every major humanoid maker uses teleop in development. The risk transfers to the reader / customer / regulator who sees a demo and assumes a shipped autonomous capability. Per DEPLOY's framework on teleoperation across manufacturers, the verified-vs-claimed framework distinguishes the three categories explicitly.

Workforce + economic displacement: real but beyond DEPLOY scope

The economic case for humanoids is straightforward. A human worker in a US warehouse costs roughly $35,000-$60,000 per year fully loaded. A humanoid leased at $5,000/month is $60,000/year and is moving toward $20,000/year as the unit economics improve. Once the lease cost crosses the loaded labor cost (which by industry projections happens during the late 2020s) the economic decision tilts toward humanoids for repetitive manipulation work. The risks here are real and beyond DEPLOY's editorial scope to resolve: displacement of warehouse + factory + logistics workers; wage compression in adjacent roles; a transition period before retraining and labor-policy frameworks catch up. For the broader cohort-wide workforce displacement question, see can humanoid robots replace workers.

Regulatory gaps: no federal humanoid-specific safety standard exists

Most workplace safety regulation predates general-purpose mobile manipulators operating in human-shared spaces. Specific gaps in 2026: no federal humanoid-specific safety standard (OSHA's existing rules cover industrial robots in cages; they don't cleanly address a bipedal humanoid walking next to a worker); liability allocation between operator + integrator + maker is unclear when a humanoid causes an incident; cross-jurisdictional variation at state and city levels is wide; no certification equivalent of automotive crash-test or FAA airworthiness for humanoids. Per DEPLOY's framework, the absence is editorial signal at the regulatory-gap layer, not weakness of the cohort.

Bottom-line risk: mis-characterizing capability

The biggest risk in 2026 is not the robots themselves. It is mis-characterizing what they can do. A teleoperated demo is not an autonomous capability. A factory pilot is not a commercial product. A press release is not a deployment. The DEPLOY bar is to keep those distinctions sharp. Per the verified-vs-claimed framework, capability claims should be evaluated against shipped-product verification depth, not marketing-layer framing. For consumer purchase evaluation, the right question is "what is verifiably shipping at consumer-deployment scale" not "what is being demonstrated."