When cyber-capable frontier models are treated like controlled strategic assets, access to top-tier intelligence starts to resemble passing through a national border checkpoint.
The strongest models disappear from ordinary cloud menus and move into sealed facilities run by approved operators. Universities, startups, hospitals, and local governments must request time inside supervised model zones, often through long queues and narrow usage contracts. Cross-border collaboration slows because knowledge work now triggers export reviews, audit trails, and diplomatic suspicion. The public still hears about astonishing machine intelligence, but encounters it only through diluted downstream services. A class system emerges: sovereign access, institutional access, and everyone else.
At 9:15 p.m. in a municipal water office in Phoenix, Elena waits for clearance to submit a leak-detection query to the state enclave, knowing the answer will come after the overnight shift has already finished the manual inspection.
Supporters argue that the bottleneck is justified. Limiting direct access to dangerous models may prevent automated cyber campaigns that no civilian institution could withstand. Yet the same barrier also concentrates capability in a handful of governments and contractors, making safety inseparable from dependence.