Bayeswatch: 2021 sci-fi story imagined AI treaty, now echoed in AI 2027

In 2021, the author published Bayeswatch, a work of science fiction intended to explore how the AI future might unfold under constraints imposed by international agreement. The story's central premise held that solving the alignment problem—ensuring that powerful AI systems behave in line with human intentions—would require major governments to coordinate and actively suppress the development of the most powerful AI systems.

At the time of writing, this framing represented a distinctive perspective on the AI safety challenge. The broader AI alignment discussion in 2021 remained largely theoretical, focused on abstract problems and conceptual frameworks. Technical work aimed at solving alignment challenges had begun, but was not yet the dominant mode of engagement. Policy work, by contrast, was tertiary in the conversation and itself largely theoretical rather than concrete. Into this landscape came Bayeswatch, advocating for a governance-based approach to a problem the field was still learning to articulate.

Bayeswatch did not remain alone in this vision. Last year, in 2025, a team of forecasters published AI 2027, which also uses science fiction to explore how an international treaty limiting powerful AI development could shape outcomes. AI 2027 has become the most popular story of this genre, suggesting that the governance-focused framing Bayeswatch introduced has gained considerable traction. The rise of these two narratives—one from 2021, one from 2025—traces a shift in how the AI research and policy communities understand the path forward: from a field dominated by theoretical alignment work to one where international coordination and policy intervention are increasingly seen as central to managing AI risks.