Figure AI sends retired F.02 humanoid robots into molten steel furnace

By Splitfeed EditorialLast updated Oct 2, 2026, 00:10 UTC

Figure AI decommissioned its F.02 humanoid robot fleet by directing the units to walk or leap into a 75-ton molten steel furnace located in Finland, with reporting on the event appearing between September 29 and October 2, 2026. The disposal was not simply logistical: the company choreographed the robots' final task as a live demonstration of their movement capabilities, having the machines perform the act of entering the furnace themselves rather than being scrapped through conventional industrial means. Separately, a report dated September 29 noted that Figure AI had identified what it describes as a robot scaling law, with a 56 percent success rate attached to the finding, though the specifics of what that law governs and how it was measured were not detailed in the available sources.

The story sits at the intersection of two developments that carry weight independently. A generational transition in Figure AI's humanoid fleet signals that the company considers the F.02 line obsolete enough to retire in bulk, which implies a successor platform is either ready or close to it. The scaling law claim, modest as a 56 percent success rate may sound, matters because scaling laws in AI have historically served as the theoretical foundation for large investment and development decisions, and establishing an equivalent for physical robots would be a meaningful step toward predicting how robotic capability improves with more data or compute. Whether 56 percent constitutes a genuine and reproducible law, or a preliminary result dressed in ambitious language, is a real and open question.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, had published reactions in the sources available. Pro-AI voices would typically frame the furnace event as a confident signal of rapid iteration and the scaling law as evidence that robotics is entering the same compounding-progress dynamic that transformed large language models. Anti-AI voices would likely raise concerns about the resource cost and theatricality of the disposal, and treat a 56 percent success rate as a reminder that humanoid robots remain far from reliable. Middle Ground observers would probably note that retiring one generation cleanly is standard engineering practice, while cautioning that a scaling law claim deserves scrutiny before being treated as a roadmap.

The clearest next signal to watch is whatever Figure AI announces as the F.02's successor, since the specifications and timeline of that platform would show whether the scaling law finding is already shaping the next design. Further detail on the methodology behind the 56 percent figure would also determine whether that result holds up as a predictive tool or fades as a marketing framing.