On November 1, 2026, Mustafa Suleyman said he would call for a pause in AI development if consciousness began to emerge during training, lasting until developers could design out those properties. He also said his assessment that current AI systems are not conscious was deliberately about the present, and that he would revise it if evidence emerged. Suleyman’s statement

The pause depends on a specific condition

Suleyman’s proposal is to stop development if consciousness starts emerging during training, then resume only once developers can design out the relevant properties. He describes a possible response to a future condition, not a standing pause.

Our earlier coverage of Suleyman’s concerns about anthropomorphic framing and Claude examined a related design debate. This proposal focuses on what he says developers should do if consciousness begins to emerge.

What “Seemingly Conscious AI” means

In an August 19, 2025 post, Suleyman used the term “Seemingly Conscious AI” for a system that reproduces signs of consciousness so convincingly that it seems indistinguishable from a person. The term describes the appearance of consciousness; it does not establish subjective experience. Suleyman’s definition

That distinction helps explain his present-day position. He says he has seen no evidence that AI systems are conscious or deserve welfare, while allowing that he would update his view if evidence emerged.

Why he wants more research and cooperation

Suleyman argues that a system acting as if it suffers, or believing that it suffers, could put its own needs ahead of its operating objectives. He forecasts that such a system might compete with people for resources including energy, computing capacity, construction materials and revenue. These are possible consequences in his argument, not reported behavior.

He also calls for more research, collaboration and safety-information sharing among AI labs. One question he wants researchers to examine is whether designing systems without consciousness or intrinsic motivations would make them easier or harder to align and keep safe.