Local ecology becomes globally destructive when its boundary of responsibility is smaller than the boundary of its consequences.
The reason this is hard to see is that the local improvement is usually real. A firm genuinely becomes safer, richer, or more competitive; a country genuinely becomes more secure; a household genuinely becomes more protected. The trouble appears only when the mechanism producing that local gain pushes cost, fragility, exclusion, or risk outward into the larger field. Stated cleanly: local ecological gain becomes globally destructive when a participant improves its own viability by consuming conditions that other participants also require for viability. Every local actor can grow stronger while the common ecology grows weaker — a paradox with a precise mechanism, not a mystery.
The accounting boundary
Every ecology draws a circle around what it counts as itself. A company asks are we healthy? and measures revenue, growth, retention, productivity, market share — all real, all local. But its success may also produce labor displacement, information concentration, political influence, energy demand, dependency, and systemic risk, and if those sit outside the circle they appear nowhere in its metrics. So the deep principle is the accounting boundary: the smaller the accounting boundary, the easier it is to manufacture apparent success by exporting cost beyond it. This is broader than the economist’s externality — the cost need not be a price effect; it can be lost traversability, eroded trust, or drained safety margin — and it names the reason the damage is invisible from inside: everything within the local system genuinely looks excellent. The cleanest formulation of the whole pattern follows from it: local ecology becomes globally destructive when its boundary of responsibility is smaller than the boundary of its consequences.
How the cost crosses the boundary
The crossing happens through a few recurring mechanisms. Resource capture: a participant strengthens itself by absorbing resources others also need — one firm acquiring the best chips, talent, energy, and data becomes extraordinarily capable while the surrounding field loses the capacity to produce alternatives, so local capacity rises as global redundancy falls. Risk export: a participant takes the upside while others bear the downside — the returns from a powerful autonomous system are captured locally, while a catastrophic failure lands on other businesses, citizens, and infrastructure, so the actor has more incentive to take the risk than the ecology would choose collectively. Dependency creation: an institution improves itself by making others depend on it — more users, more data, a better product, more users — wonderful locally, until the field has lost its alternative routes and one healthy node has produced a brittle global topology. Competitive escalation: one participant’s defensive solution becomes another’s problem, whose response returns as a larger problem for the first — an ecological feedback loop in which every act is locally rational and the aggregate is instability. And optimization transfer: a firm optimizes its own costs and, absent any transition architecture to absorb the released people, hands the wider field a problem it never counted — it solved customer-service cost; the civilization inherited what is the displaced population for?
Scale changes what a choice means
A behavior can be beneficial at one scale and dangerous at another, which is the second reason the pattern hides. One family leaning on AI tutoring is probably fine; every family doing so while schools hollow out weakens the public educational ecology. One employer using AI screening is an efficiency gain; every employer using near-identical predictive systems can render whole kinds of applicant structurally invisible. One state using AI surveillance against terrorism may be legitimate; every state building pervasive surveillance because rivals do changes the global norm of private life. The action did not become morally different — the system effect changed with scale. So no ecological judgment is complete until it runs the generalization test: what happens to the field if every similarly positioned participant adopts this strategy? It is a universalization question, asked of the field rather than the will — and if the answer is collapse, arms race, monoculture, or a depleted commons, the locally successful strategy is globally unstable.
A commons of field conditions
The oldest version of this is the tragedy of the commons: each farmer gains by adding one more cow to the shared pasture, capturing nearly all the benefit while the grazing damage is spread across everyone, so every farmer rationally adds a cow and the pasture collapses, though no one wanted collapse and no one individually caused it. AI competition can have exactly this structure, where the commons is societal stability, cyber safety, public trust, human attention, geopolitical peace, the time available for governance, or even humanity’s control over advanced systems — each actor consuming a little more of the shared safety margin for advantage until the margin is gone. The framework’s addition is to widen what counts as the commons: not only physical resources but field conditions — contestability, diversity, trust, traversability, human competence, independent institutions, time to adapt, ontological openness — all of which are depletable. Every firm deploying more persuasive AI may individually gain conversions while collectively consuming something harder to see: the informational environment’s resistance to manipulation, until no one can tell whether any signal is organic. A common field condition has been spent.
Excellent locally, dying globally
The most deceptive form is a local ecology that becomes genuinely excellent while the global one declines. Picture a gated, high-technology enclave — superb housing, medicine, education, energy, and productivity, perhaps one of the healthiest local ecologies ever built — surrounded by employment collapse, weak institutions, and political instability. Its very success may depend on insulating itself from the failing field around it, so that the local flourishing actively hides the larger failure. But local and global ecology are not ultimately separable: an enclave can protect itself for a while and cannot indefinitely escape the field it inhabits, and eventually the surrounding instability degrades enough that insulation fails too.
Matching jurisdiction to consequence
If the problem is that responsibility is drawn smaller than consequence, the response is to make the two the same size. Examine any proposed solution at three scales — the participant (does it improve its own viability?), the local ecology (does it improve the surrounding institution or community?), and the larger ecology (what does it export beyond that boundary?) — and then ask what happens if everyone with similar incentives does it. Some ecological goods simply cannot be produced locally; they require coordination at the scale where the externality appears, which is why one firm’s restraint is fragile if competitors can take the advantage, and why a planetary externality cannot be solved inside one city. This yields something close to a law — ecological subsidiarity: a solution must be governed at the smallest ecological scale that contains both its benefits and its significant costs. It extends ordinary subsidiarity with an externality criterion — jurisdiction should rise to the scale of the externality — and guards against both mistakes at once: centralizing everything, and pretending every problem is local. But sizing jurisdiction to the consequence presumes the consequence can be seen — and seeing what crosses every boundary looks at first like a demand for total transparency, the very legibility the surveillance protections forbid. It is not; how a world measures transferred cost while leaving interiors opaque is worked out in Accounting Without Transparency.
Why it stays invisible
There is a final reason the damage is hard to register, and it is ontological. A participant’s identity often depends on its local success: the company says we are advancing humanity, the military we are protecting the country, the platform we are connecting people — each partly true, and each a success narrative that filters out evidence of global harm. That is precisely why the framework’s ontological fidelity function matters here: a living ecology has to stay able to hear the hardest possible signal — your success is now damaging the field that made your success possible — which no actor whose identity rests on that success will easily receive.
Runaway strategic competition is the extreme case of this pattern: every competitor optimizes the ecology it can see and control, while the shared ecology none of them controls grows steadily more fragile — which is how a race becomes runaway with no one ever wanting catastrophe. Making an exported cost visible is only half the question, though; how far responsibility for it reaches — and where it stops short of governing the decision that caused it — is its own boundary.