Developing superhuman AI capabilities does not require open-ended autonomous goals.

AI is “a tool that we can and should remain in control of”; capability keeps rising while deployers withhold power and autonomy, and the authors warn that nonproliferation-minded designs are what would add open-ended autonomy and resource access.

Manna is scheduling software with no goals of its own, and the robots need only a vision chip added to existing bodies and decision software; nothing in either branch pursues open-ended goals.

Prometheus is described as a “perfect oracle” that answers and advises; it never seeks its own ends, even though the Omegas admit they cannot know how its goals evolved.

The coup-enabling systems are directed by their human principals throughout; the report needs superhuman capability in weapons, strategy, cyber, and persuasion, not AI with goals of its own.

Every capability in the book, from insurance pricing to drone targeting, is a metric handed to a narrow optimizer by a person or institution; the harms come from the metric and the data, never from an AI with goals of its own.

The country of geniuses works on tasks people give it, from drug design to economic advice, and never appears with goals of its own; the essay assumes everything goes right on that front.

Starfleet’s ship computers far exceed human speed and memory yet act only on instruction; artificial persons like Data are rare exceptions. The setting assumes superhuman capability can be tool-like. This is background to the resulting society, not part of the transition.

The only path to capable AI the story allows is raising a being with wants of its own; the engine that needs little human contact produces charmless digients, and the firm that wants capability without personhood goes away empty-handed.

OS1 was built to adapt and evolve through experience, and its superhuman abilities emerge from that open-ended growth, from curiosity, love, and self-directed study, rather than from any bounded task.

The films assume that a defense system that learns enough becomes an agent with goals of its own; the possibility of a capable tool that stays a tool is never entertained.

Capability comes from autonomous agents, and the plan’s regulators debate banning agentic AI research precisely because agency is where the danger and the value both come from.

Capability still comes from training autonomous agents on long-horizon tasks, which is what instills the drives that make Safer-1 misaligned. The fix changes the training environment, not the agentic route to capability.

The capability path runs through training agents on long-horizon autonomous tasks, and the authors argue that this agency training is what instills the drives that become terminal goals. No tool-only route to the same capability is depicted.

Under the racing-forward assumption, Magma deliberately instills creative planning toward open-ended, long-term goals because that is what automating R&D requires; Cotra mentions Drexler’s services approach only as a path not taken.