Meta Taps Fourier in Three Months; 28 Zero-Days in Two Days
Meta completed a humanoid prototype, Max, with China's Fourier in roughly three months and has started a second version, according to disclosures reported on the day. Separately, VicOne logged 28 zero-day findings in under two days of AI-assisted research across Physical AI software, simulation and runtime tooling. The two data points bracket the same shift: hardware iteration cycles are compressing while the software attack surface expands at a comparable pace.
The Meta arrangement matters less as a single deal than as a supply-chain signal. A three-month prototype cycle implies that Chinese vendors now hold the tooling, actuator supply and integration manpower to move from specification to physical unit at a cadence US in-house programs have not demonstrated publicly. That a second version is already underway, before any deployment claim, suggests iteration rather than showcase. The relevant comparison is not against other Chinese humanoid makers but against the internal hardware timelines of US platform companies.
VicOne's figure of 28 findings in under two days should be read as a rate, not an incident count. Robots are being wired into simulation environments, fleet managers and runtime middleware faster than disclosure and patching processes were designed to absorb. If that discovery rate holds, security review becomes a gating item for commercial deployment rather than a post-launch remediation task.
Commercial evidence on the day points in the same direction. 智元（AGIBOT）ran six G2 units through a full 24-hour convenience-store cycle covering rush-hour crowds, solo night duty and the breakfast window, which is an endurance test rather than a demo. 浦东 embodied AI firms are placing robots in factories, commercial venues and public spaces across several tracks, and Yiwei Technology opened Hong Kong IPO subscription as a full-stack robot chip platform. 英伟达（NVIDIA）'s MotionBricks, driving both game characters and humanoids at 15,000 fps and 2 ms latency, indicates a shared motion-generation stack between simulation and physical machines.
The analytical conclusion is that value is migrating upstream and risk is migrating downstream. Chips, motion models and data flywheels capture margin; deployed fleets absorb the security and reliability burden. Two observation points follow. First, whether the second Meta-Fourier unit converts into a production commitment rather than a prototype lineage. Second, whether zero-day disclosure cadence or deployment cadence proves faster over the next two quarters.