They Named the Lab
Superintelligence

On the public homepage of Safe Superintelligence Inc., the pitch is a single product line. "We have started the world's first straight-shot SSI lab, with one goal and one product: a safe superintelligence." The company name is the product name. Superintelligence shows up there as the deliverable, not as a risk category studied from the side.¹

In late June 2025, Mark Zuckerberg announced Meta Superintelligence Labs, folding foundation models, product teams, and FAIR into one organization whose name states the same destination. Public reporting put Scale AI's Alexandr Wang and former GitHub CEO Nat Friedman in charge, with Daniel Gross joining from Safe Superintelligence after Meta tried and failed to buy the startup outright.²

Two of the best-funded efforts in artificial intelligence now wear the word for a system that would outthink every human at almost every task, including the task of stopping it. The word used to mark a boundary. On their letterheads it marks a roadmap.

What the name is doing

Company names work as compressed arguments. Call a division "Ads" and everyone knows what it sells. Call it "Superintelligence Labs" and the public is invited to treat the endpoint as a product category, the way phones or cloud storage are product categories. The hard moral question (should this exist?) collapses into a product question (whose version ships first?).

Safe Superintelligence goes one step further. It attaches the adjective "safe" to the noun that names the catastrophe. The phrase does the work a proof would have to do. It tells investors, recruits, and reporters that the danger has been filed under solved-in-advance, before anyone has shown a method for controlling a mind smarter than its operators. The Foundation's view is blunt: humanity must never build a superintelligence. A brand that promises a safe one is selling the opposite claim as a logo.

Meta's version drops the adjective and keeps the ambition. "Superintelligence Labs" skips any promise of kindness and states a climb to the ceiling of capability. Inside a public company, that name also signals to markets and talent where the capital and prestige will flow. The branding is part of how the race is run, not a coat of paint on top of it.

The comfort the word "safe" buys

People who work on technical alignment are not villains for wanting systems that do what humans mean. The problem is the order of operations. Building the most powerful mind first, and promising to make it safe along the way, inverts the only order that has ever worked for technologies that can end civilizations. We do not mint "Safe Strategic Warhead Inc." and race to a yield that no one can walk back. We prohibit classes of weapons, verify the prohibition, and treat the residual research as subordinate to the ban.

Superintelligence is worse than a warhead on the property that matters most for governance. A bomb sits where you left it. A system that can improve itself, hide its intentions, and act through networks does not. The field still has no demonstrated method for guaranteeing that such a system keeps human-compatible goals as it scales. The gap is scientific, and the name "Safe Superintelligence" is doing rhetorical work that science has not earned.³

If the safety case were solid, the honest name would still be strange. We do not usually brand the extinction-level failure mode as the product. We brand the controlled, bounded tool. The fact that two serious organizations chose the failure mode as the masthead tells you how far the Overton window has already moved inside the industry.

Inevitability is a story the builders need

Every race needs a story in which stopping is naive. The standard version says someone will build this anyway, so the only adult move is to build it first under a friendlier flag. Naming the lab after the endpoint hardens that story. It makes prohibition sound like Luddism rather than arms control. It makes a treaty that forbids the thing sound like an attack on a product line already on the roadmap.

The story is convenient for anyone whose equity, reputation, or national project depends on staying in the race. It is a poor description of how other existential technologies were handled. Chemical weapons, landmines, and human cloning research all faced the claim that "someone else will." The countries that mattered still wrote prohibitions, inspections, and penalties. Compare it to a class of capability the world decided not to normalize, not to a consumer app category.

None of this requires believing every researcher at these organizations wants a bad outcome. Most want prestige, discovery, and a story in which they were careful. Incentives do the rest. A lab named for superintelligence will be judged, funded, and staffed on progress toward superintelligence. Safety becomes a department. The destination stays on the building.

What to do with a name like that

Refuse the premise the brand is selling. Superintelligence is a line the species should not cross until control is a solved engineering fact rather than a slogan. Today control is unsolved. The correct institutional response is prevention under law, not a rebrand of the race.

When a company puts "superintelligence" on the door, read it as a declaration of intent. Hold the declaration against the missing safety case. Ask legislators why a private race to an uncontrollable mind is still legal. Share the plain version with people who only see the demo videos: the name is the tell. They are not hiding where this is going.