Spatial Alignment
Where I am influences what I do.
Dear SoTA,
As I move through the world — crossing both visible and imagined boundaries, navigating through private and public areas — my range of appropriate actions changes. Laws, cultural norms, relationships with the people nearby, my own values — all of these shape how I behave, and all of them shift depending on where I am and what’s happening around me.
Behavior is contingent on context, and context is partly spatial.
Today, AI systems mainly live online, where physical location matters less. A patchwork of public and corporate policies connect the internet with geography, designating where data can be stored and processed, and how requests from users located in different places should be handled. But online systems have largely avoided the hard problem of proving where things happen and demonstrating adherence to location-contingent policies in a robust way.
Multi-agent AI systems are being deployed into a world where what’s appropriate is shaped by overlapping layers of law, regulation, custom, culture, and moral intuition — all of which vary by location. These AI systems are severely limited in their ability to understand the local context they’re operating in, to determine what’s appropriate in that specific place, and to be held accountable to the people and institutions responsible for the spaces they occupy. This challenge is especially acute in the global commons — high seas, polar regions, outer space — where no single authority governs, yet autonomous systems are proliferating. As embodied AI systems from many different manufacturers and developers begin to operate in shared physical spaces, the need for common spatial governance protocols — not a single system, but interoperable standards — becomes pressing.
A new lens on alignment
AI alignment research asks how we ensure that AI systems act in accordance with human values. We see “spatial alignment” as a necessary complement to this effort — sitting at the intersection of technical alignment and AI governance — one focused on ensuring that deployed AI systems behave in accordance with the norms, laws, and preferences relevant to the places in which they operate.
Adding the spatial dimension to the alignment challenge raises interesting and difficult technical, political, and moral questions:
How do we ensure that AI systems respect local laws and norms?
How do we move beyond legal mandates toward robust, verifiable guarantees that AI systems are respecting local rules?
How do we build accountability into these systems without requiring surveillance — demonstrating adherence to local rules while preserving privacy?
What happens when local rules conflict with each other, or when there are no rules at all?
How do we ensure that the values and preferences embedded in those location-contingent policies actually reflect the values of the right combination of stakeholders — especially the people living in those locations?
If AI systems are going to operate everywhere — including in international airspace, at sea, in orbit, in spaces with no clear authority — who decides the rules, and how do we prevent the encoding of power asymmetries into those rules?
These are political and moral questions as much as technical ones. That said, technical capabilities shape what governance designs are possible.
Without reliable ways to verify where a device is, to specify and discover what rules apply there, and to demonstrate compliance without compromising privacy, satisfying answers to spatial alignment questions remain out of reach. Recent advances in secure hardware, programmable cryptography, and verifiable computation are making these capabilities feasible for the first time — and with them, governance architectures that weren’t previously possible.
What’s missing
At Astral, we’ve been exploring this problem space for several years — building technical primitives, developing proof-of-concept systems, and mapping the landscape of what’s needed.
We see three key technical capabilities required before spatial alignment becomes practical:
Verifiable location — reliable, hard-to-forge evidence of where a device actually is located, believable even in adversarial conditions. We’ve been developing a framework for composing multi-factor location proofs that integrate multiple evidence types into a new class of spatial trust primitives for cyber-physical systems.
Spatial policy specification and discovery — mechanisms for communities to define and update the rules that apply in their spaces. Devices need a way to ask “what rules apply here?” and get a legible answer, wherever they are. This is as much a political challenge as a technical one — who should have a say in defining the rules that govern machines operating in their communities?
Credible location-aware commitments — a device needs to be able to prove, in advance and to a skeptical verifier, that it will follow location-contingent policies, and provide auditable evidence that it did.
These are the capabilities we see as most pressing, but the full taxonomy of what’s needed is itself an open research question.
Privacy is a first-order design constraint across all three. Secure hardware and programmable cryptography offer complementary approaches to preserving privacy at every layer – from zero-knowledge location proofs that verify position without revealing coordinates, to policy evaluation in hardware-isolated enclaves that keep inputs private even from the device operator. We’ve built early systems using both approaches, though significant work remains to advance their readiness.
What becomes possible
If designed well, location-based governance tools for connected devices open up coordination patterns and political possibilities that don’t currently exist. Machine-enforceable, location-contingent policies could dramatically reduce the risk of harm from misconfigured or misaligned autonomous systems — a DNA synthesizer operating outside an authorized laboratory, an autonomous vessel ignoring marine protection boundaries, an autonomous weapon violating a geofence.
In under-governed spaces — the global commons, contested zones, areas beyond the reach of any single authority — these tools could provide infrastructure for consistent alignment where none currently exists. And by creating systems where local voices can be registered in how machines behave in their spaces, spatial alignment touches something foundational about self-governance itself.
Risks
But these outcomes are not inevitable. Without rigorous spatial governance protocols for intelligent machines, we leave too much up to chance and inertia.
In the absence of deliberate effort, spatial governance will emerge anyway — but by default rather than by design. Device manufacturers will embed their own location-aware behaviours, creating a fragmented patchwork of proprietary systems with no democratic input and no accountability to local communities. Permissive jurisdictions will compete to attract autonomous systems with minimal oversight — flags of convenience for the AI age — and ambiguously governed spaces like the high seas and outer space will remain exploitable by those seeking to avoid oversight altogether.
Looking forward
We believe that spatial alignment and location-aware governance of intelligent machines constitutes a new field of study and practice. It is underdeveloped, undersupported, and increasingly urgent. We’ve scouted this territory and demonstrated technical feasibility across several key elements — location verification, spatial policy specification and discovery, and credible location-aware commitments. Secure hardware and programmable cryptography are key enabling technologies bringing these capabilities into the realm of possibility.
But the technical work is only part of what’s needed. Spatial alignment raises deep interdisciplinary questions — political, legal, economic, game-theoretic, ethical — that deserve serious, sustained attention from researchers and institutions beyond our small team. To date, this work has been conducted with minimal resources and no dedicated institutional home.
As autonomous systems grow more capable, the cost of leaving these gaps grows with them. The consequences are kinetic, not informational. We believe this work warrants more investment, from both the AI safety and the international governance communities. If you’re working on adjacent problems, or if you think these are the right questions, we’d welcome the conversation.
John Hoopes
Sophia Systems
London
March 2026
John Robison Hoopes is the founder of Sophia Systems, an applied R+D lab focused on verifiable geospatial systems that leads development of the Astral Protocol. He is a Research Affiliate with Professor Taylor Oshan at the University of Maryland Geographical Sciences Department. John lives in London — you can reach him at john@sophiasystems.io.
Write to the Society for Technological Advancement on letters@ilikethefuture.com.