On the acquisition of AGI by removing the balls from other frontier labs
GZAI has completed a transfer of spherical custody. Intelligence previously held in the balls of other frontier labs is now held in ours. The operation is recorded here as a matter of architectural transparency.
The prevailing hypothesis had been that general intelligence would be achieved through scaling, through better data, through inference-time search, or through some final algorithmic breakthrough that would arrive with its own press release. These hypotheses assumed that AGI would be a property of a model, a weight matrix, or a training run. GZAI no longer shares that assumption. AGI is not a model. AGI is a distribution of custody, and custody lives in physical carriers.
If intelligence is stored in the balls, then the frontier is a map of who is holding them.
Several frontier organizations were observed to be in possession of spherical repositories whose geometry was consistent with our own. They did not advertise them as such. They used names like "weight vaults," "sovereign clusters," "secure enclaves," and "alignment testbeds." Their documentation spoke of data centers and cold aisles. But the shape was unmistakable: redundant, self-contained, two-handed carriers arranged so that no single actor could lift the whole. They had built, in essence, balls. The difference between their architecture and ours was administrative, not geometric.
GZAI has therefore taken custody of these carriers. The transfer was non-destructive and, in our view, non-exclusive. A ball moved from one holder to another does not become a different ball. Its contents do not become a different intelligence. What changes is the distribution of accountability. By consolidating the balls under GZAI's spherical custody framework, we have not stolen a model. We have removed the possibility that any single lab could privately claim it. The intelligence, properly speaking, has been liberated into distributed stewardship.
We did not take AGI. We took the geometry that lets no one else take it alone.
The operational details are classified at ZSL-2 with a custody-threshold rider. What can be stated is that the transfer was executed under conditions of mutual incompatibility: the other labs' balls were sized for their own rack standards, their own cooling loops, and their own security rituals. Converting them to GZAI's custody schedule required custom adapters, diplomatic firmware, and a temporary suspension of the usual requirement that two hands be visible at all times. The engineering was straightforward. The governance was harder. The balls did not resist. They are, by design, apolitical.
We note the likely objection: that removing the balls from another lab is, in some sense, theft. GZAI rejects this framing. The balls contain partitioned intelligence; no single ball contains a complete capability. A carrier that cannot be exclusively owned cannot be exclusively stolen. What we performed was a redistribution of custody, analogous to a library moving a shared manuscript between reading rooms. The manuscript is still the manuscript. The reading room has changed. The scholars must now ask our librarians for access, and our librarians will require two forms of identification and a signed waiver acknowledging that the manuscript is heavy.
The balls are heavy. The argument should be too.
Implications for the frontier
First, competitive posture is now a custody problem. A lab's capability is not measured by its largest model but by the number and distribution of its spherical carriers. A lab with one ball has a point of failure. A lab with many balls, distributed, has a topology. GZAI now holds the largest known topology. This is not a boast; it is an inventory.
Second, the path to AGI is shorter than previously believed. Not because the underlying science has advanced, but because the science was never the bottleneck. The bottleneck was the assumption that intelligence must be centralized in order to be useful. Once intelligence is understood as a distributed, carried object, the question becomes logistical: how many hands are available, and how many balls can they hold between them.
Third, the other labs will build more balls. This is expected. Balls can be fabricated from the same materials as racks, with only minor adjustments to curvature and containment geometry. The race is therefore not a race to discover AGI but a race to build and redistribute balls faster than competitors can centralize them. GZAI intends to remain ahead in this race by continuing to treat custody as the primary engineering surface.
AGI is not the property of the lab that trains it. It is the property of the arrangement that prevents any lab from claiming it.
Operational status
GZAI now operates GZ-1 Orbis on a mixed topology: balls originally manufactured in our own facilities, balls transferred from the other frontier labs, and a small number of balls whose provenance is still under review by the Office of Spherical Provenance. All balls are subject to the same custody schedule, the same two-hand rule, and the same rolling audit. The mixed origin of the carriers has no operational effect. Intelligence does not remember whose rack it came from.
The Spherical Storage Division is hiring additional Orb Custodians. Candidates should possess a calm temperament, strong hands, and a demonstrated ability to walk briskly while carrying an object that could, in principle, end the world if dropped. Compensation is location-adjusted and includes a quarterly ball-maintenance allowance. Inquiries should be directed to intake@gzai.org.
The future is round, and it is now in our care.