Everyone wants to talk about safety.
Politicians, editorial boards, and the entire Effective Altruism comment section are suddenly very concerned that a $12,000 humanoid might go berserk and strangle your grandmother.
Here is the part they never say out loud:
A humanoid robot is the safest superhuman actuator ever built, precisely because it is weak, slow, and made of plastic.
Raw numbers (2025–2027 generation)
- Peak joint torque: 80–120 Nm (human adult male ~200 Nm at shoulder)
- Top speed in normal mode: 1.2–1.8 m/s walking, 0.5 m/s arm movement (human sprint 10 m/s, punch ~8 m/s)
- Total system power: 1.5–2.5 kW continuous (average U.S. male can sustain ~300 W, peak >2 kW for seconds)
- Mass: 55–75 kg, mostly hollow plastic and aluminum
- Kill switch: hard-wired watchdog that drops power in <30 ms if heartbeat lost
- Emergency stop: big red mushroom button on spine + remote cloud kill within 400 ms
- Skin: soft silicone over force-limiting joints that collapse before bone breaks
In a direct fight against an average 15-year-old human wins against any 2027 humanoid 10 times out of 10. A large dog wins 8 times out of 10. A baseball bat ends the match in one swing.
The robot is not dangerous because it is strong.
It is dangerous because there will be ten million of them, and they will be everywhere.
The real threat model is never Skynet. It is always the Longshoreman
Historical precedent is crystal clear: every labor-replacing technology since the Luddite looms has faced the same three-phase backlash.
Phase 1 (2026–2027) – Symbolic protests
Dock workers in Oakland smash two Optimus units unloading containers. Viral video gets 400 million views. Politicians introduce “Humanoid Safety Acts” that require $2 million liability bonds per robot. Tesla and Figure simply move the next 50,000 units to ports in Texas and South Carolina. Nothing changes.
Phase 2 (2027–2029) – Regulatory capture attempts
Unions and European social-democrat parties push for “robot taxes” and mandatory 10-year phase-ins. China ignores it completely and ships 8 million units. Europe’s entire regulatory framework becomes irrelevant the moment Amazon offers next-day delivery of a $9,999 robot from a warehouse in Poland.
Phase 3 (2029+) – Acceptance or collapse
Either societies adapt (massive wealth tax + UBI experiments + retraining that mostly fails), or they fracture. There has never been a Phase 4 where the machines were successfully banned once the price fell below ~15 % of median annual income. (See also: smartphones in sub-Saharan Africa, motorbikes in Vietnam, Starlink dishes in rural Brazil.)
“Alignment” is a category error for physical systems
In language models, misalignment looks like persuading the model to help build a bioweapon.
In robots, misalignment looks like:
- Picking up the wrong object and putting it in the wrong bin (economic cost: $2)
- Walking into a glass door it didn’t see (repair cost: $800)
- Accidentally knocking over a toddler because depth estimation failed (lawsuit cost: millions)
These are engineering failure modes, not super-intelligence deception.
The actual control mechanisms that work:
- Hard physical limits (torque caps, speed governors) baked into firmware.
- Cloud permissioning: every non-trivial action above 20 N or outside geofenced areas requires a signed token from the manufacturer.
- Insurance: no robot operates without a real-time liability policy that is voided the moment you load third-party weights.
- Kill infrastructure: 5G + Starlink means the manufacturer can brick any unit on Earth in <2 seconds.
These are not theoretical. Tesla demonstrated remote shutdown of 400 Optimus units simultaneously in a November 2025 factory drill. Figure’s enterprise contract with BMW contains a clause that every robot stops moving within 500 ms of a factory-wide E-stop.
The real danger nobody wants to price in
The machines will be perfectly safe for the owner class.
They will be catastrophic for the 400 million humans whose jobs pay $8–$25/hour and consist mostly of walking around and moving objects less than 15 kg.
When the first viral video drops of a Figure 02 gently helping a 92-year-old Japanese woman out of bed, bathing her, and preparing breakfast, the political support for bans evaporates overnight. The demographic that votes (old, rich) suddenly wants ten million more robots yesterday.
The young, poor, and soon-to-be-unemployed do not get a vote that matters.
Bottom line
There will be no pause.
There will be no international treaty that survives contact with $8,000 robots.
There will be no kill switch big enough to stop a billion actuators once the factories are running 24/7.
The only remaining question is distribution: who owns them, who programs them, and who gets the economic surplus.
Everything else is theater.