The Specific Event

California Governor Gavin Newsom signed an executive order this week directing state agencies to examine the feasibility of kill switches for AI models, along with other regulatory mechanisms he had previously vetoed in legislative form. The move is notable precisely because of that history: Newsom blocked SB 1047 in 2024, citing concerns that broad AI safety legislation would stifle innovation. Now, facing a potential 2028 presidential run, he is advancing similar ideas through executive action rather than statute. The policy substance matters less here than the structural pattern it reveals. A state government is attempting to impose top-down control mechanisms on systems whose operational logic is not fully legible to the regulators designing the controls.

Kill Switches Assume You Know What You Are Killing

The conceptual problem with a kill switch for an AI model is not technical. It is epistemological. A kill switch presupposes that the system has a discrete, identifiable output that can be halted cleanly. But agentic AI systems, and increasingly the algorithmic management systems Kellogg, Valentine, and Christin (2020) describe in workplace contexts, do not operate as discrete processes. They operate as coordination mechanisms embedded in organizational workflows. Stopping the model does not stop the dependencies that have formed around it. This is the same organizational opacity problem that motivated my earlier writing on Zeroset, but the California order applies it at the regulatory level rather than the firm level.

The Awareness-Capability Gap, Applied to Regulators

The Canadian opinion piece surfacing this week makes a parallel argument about algorithmic workplace surveillance, calling on regulators to treat AI-driven management as a psychosocial hazard before harm accumulates. Both the Canadian framing and the California order reflect what I would call a regulator-level awareness-capability gap. Policymakers clearly have growing awareness that AI systems create new categories of risk. What they demonstrably lack is the structural schema needed to translate that awareness into effective intervention design. Knowing that a kill switch could exist is not the same as knowing what organizational and computational conditions would make it meaningful. This is precisely the distinction Hatano and Inagaki (1986) draw between routine expertise, knowing a procedure, and adaptive expertise, understanding the principles that make a procedure work in novel contexts.

Why Executive Orders Are Topographic, Not Topological

Using the distinction I developed in my previous post on YouTube's algorithm changes, there is a difference between topographic knowledge and topological knowledge of a constraint system. Topographic knowledge maps specific features: which model, which deployment context, which risk category. Topological knowledge understands the structural shape of how AI systems create organizational dependencies regardless of which specific system is involved. Newsom's executive order is topographic. It targets specific mechanisms like kill switches and specific risk categories without engaging the underlying structural question of how algorithmically mediated coordination changes the relationship between a governed organization and the governing authority. Rahman (2021) describes this dynamic as the invisible cage: workers inside algorithmic systems face structural constraints that are real but not directly observable. Regulators face an analogous problem from the outside.

What Newsom's Order Actually Signals for Organizational Theory

The more theoretically interesting implication here is what this event reveals about the limits of hierarchical governance applied to platform and AI coordination mechanisms. Classical organizational theory, as reflected in the coordination literature, assumes that the entity imposing control has sufficient information to design effective controls. Markets, hierarchies, and networks all depend on some version of this assumption (Kellogg et al., 2020). Algorithmic systems break this assumption because their coordination logic is endogenous: it adapts based on the behavior of participants, including regulators who interact with it through policy signals. An executive order requiring kill switch feasibility studies may itself become a signal that shapes how AI developers configure their systems in California, independent of whether any switch is ever deployed.

The Practical Question This Raises

The Canadian and Californian regulatory moves, taken together, suggest that democratic governments are converging on awareness without convergence on schema. They know something needs to be governed. They do not yet have an accurate structural model of what the governance target actually is. For organizational theorists, this is not a criticism of regulators. It is a research problem. The ALC framework I am developing asks how individuals develop transferable structural understanding of algorithmically mediated environments. The same question applies at the institutional level. What would it look like for a regulatory body to develop genuine topological literacy about AI governance rather than producing successive rounds of topographic rules that the systems themselves route around?

References

Hatano, G., and Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, and K. Hakuta (Eds.), Child development and education in Japan (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., and Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.

Rahman, K. S. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945-988.