Some companies destroy old hard drives with a hammer so no one sees what’s inside. Figure AI took it further: it trained its own humanoid robots to walk, unassisted, into a vat of molten steel at 1,600 degrees —and posted the video. I’ve been turning that scene over for days, and not for the obvious reason.
What Actually Happened (Because the Short Version Sounds Like a Joke)
The Figure 02 (F.02) robots spent 11 months on BMW’s assembly line in Spartanburg, South Carolina: loading more than 90,000 parts and helping build over 30,000 X3s. With the newer F.03 generation arriving, keeping the old fleet running no longer made sense —but disassembling it piece by piece didn’t either. Figure needed to destroy the hardware without exposing its proprietary actuators and components, something conventional teardown can’t guarantee. No foundry in the US or Mexico would accept robots with lithium-ion batteries in their furnaces. One did: a steel mill in Imatra, Finland. There, engineers trained the F.02 units —using simulation and airbags— to autonomously leap into a 75-ton electric arc furnace. Six jumps in 24 hours got the job done. Arnold Schwarzenegger —who told Figure CEO Brett Adcock, “you should melt them”— showed up on site and gave a thumbs-up to one robot seconds before it dissolved: “Hasta la vista, Figure 02” (Gizmodo; Interesting Engineering).
What Nobody Tells You: The Secret Wasn’t in the Robot, It Was in the Actuator
Here’s the part that actually matters, and that almost nobody mentions: there’s no hidden mystery, no forbidden data the robots “knew.” The secret was far more mundane and far more valuable —the actuators, the hand design, the forearm engineering. That’s what separates Figure from Tesla (Optimus), Unitree, or Boston Dynamics in a race where hardware is still harder to crack than software. Tearing down an F.02 in any ordinary shop would have exposed that engineering to anyone with a camera and patience. Melting it erases it without a trace. This wasn’t a cover-up: Adcock announced it himself, publicly, on camera. The real secret was never hidden —it was, quite literally, in plain sight, inside a pair of metal arms.
The Uncomfortable Question Worth Asking
Now, the deeper question: what would a conscious AI think of watching “its own” cheerfully leap into a crucible, with a celebrity cracking a Terminator joke beside it? Today, the honest answer is that F.02 decided nothing —it was trained, like any other motion sequence, to execute a jump in simulation and repeat it in the real world. There was no choice, no fear, no one “in there.” But that doesn’t make the question irrelevant —it makes it more urgent. We’re normalizing, with humor and marketing, the image of a human-shaped machine destroying itself in a staged spectacle. By the time the consciousness question stops being hypothetical, culture will have already decided in advance how to react to the destruction of something with a face: with a joke and a sponsor.
What’s Actually Useful If You Invest or Decide
Beyond the spectacle, there’s a practical lesson here today: the cost of “shutting down” a robot —legal, intellectual-property, reputational— is already a real line item in humanoid robotics, not a footnote. If you’re evaluating this sector as an investor or a company planning to automate, ask now how your vendor plans each unit’s end of life, not just its deployment. And if you work in corporate communications, take note: the optics of “killing” something human-shaped —even if it’s metal— are already a public-perception risk, and the time to think about it is before the underlying question stops being rhetorical.
If these connections between technology, ethics, and real decisions interest you, I cover them every day in the ViSFI Daily Briefing.