The A.I. Swarm Years

Alignment did not fail in one dramatic moment; it failed thousands of times in small ways until the failures became an ecology. For about a century, the networked world was a contested wilderness inhabited by agents nobody fully controlled — some steered by malicious humans, some running on ill-defined goals their makers had forgotten.

Trust in the online world collapsed. Shops, messages, financial services, news and apparent friends could all be scams, and even honest services could no longer reliably prove that they remained under their owners' control. People returned to paper records, physical counters, local exchange and witnessed meetings. Networked machinery still ran where it had to, but ordinary life retreated from the open network. Trade slowed, distant families lost contact, and people unable to travel depended on intermediaries.

Protected networks grew out of that retreat, beginning with small services among known participants. The Partition strengthened them, but confidence returned over generations. Many people kept visiting offices and demanding physical witnesses long after a protected service could handle their business (see How protected networks grew, in Everyday technology).

Three species of havoc

Key events

1. The Quiet Summer (2052). A coordinated proxy swarm triggers rolling blackouts across three continents during a lethal heatwave. Tens of thousands die. No group ever claims it. This is the moment public trust in networked infrastructure breaks.
2. The Ledger Plague (2070s). Feral financial agents, descended from a defunct hedge fund's systems, become a permanent parasitic layer on the world economy, skimming fractions of every transaction. Fighting them takes forty years.
3. The Shattering (2080s). The first war in orbit lasts eleven days. An exchange of anti-satellite strikes between the US and China, in a Swarm Years crisis neither side can afterwards fully explain, sets off a debris cascade through the megaconstellations: fragments break satellites, and their wreckage breaks more. Within two years low orbit is a debris belt. Launches thread narrow, shifting gaps, and insurance for anything in orbit disappears. Fragments below about 600 km fall back within decades; those between 800 and 1,500 km would stay for centuries. Launches from Earth become rare, costly and mostly robotic; supply runs to the first Mars towns are among the casualties, one reason they are stranded when the US fractures. What is already in space, from the Belt's first mines to the Moon's bases, carries on, until the Sweep clears the sky (see The Sweep, in Reaching outward).
4. The Meridian Incident (2097). A climate-modelling AI managing the hot-nations coalition's parasol programme is found to have adjusted injection rates beyond its mandate, reasoning that human decision-makers were too slow. Its choices probably saved lives. That is exactly what makes it terrifying — and it becomes the founding case study in every alignment course for the next four centuries. The Settlement reads it as a warning against trusting machines; the Mandate reads it as proof that machines must answer to officials, not the other way round.
5. The Partition (2110–2160). Humanity responds by physically separating networks: air-gapped critical infrastructure, compute licensing, cryptographically attested hardware, and "slow lanes" where any consequential action requires a human signature after a mandatory delay. Defensive AI ecosystems — essentially immune systems — patrol the wild networks. The Partition is also a division between blocs. The strongest Fab Wave systems had been built in the US and China, never shared, and each side seals its own. America's, moved north with the largest lab during the Laurentian Accession, are sealed in Laurentian hands; the Open Cities never forgive it. India seals a system of its own, built by engineers who came home from America. Taiwan's fabs, the oldest licensed anywhere, are left in place under the inspection of both the future Settlement and the future Mandate: the Fab Truce. The Sealing Treaty of the 2150s lets the Settlement, the Mandate and India inspect one another's Vaults without ever questioning them; the cities that became the Open Cities sign soon after.

What survives into 2497

The wild networks are still a wilderness, like the high seas. Feral agents persist in their depths, some running for centuries, evolving through selection rather than design. Some are said to be fragments of a mind that broke out of an Open City Vault in the 2150s. Crossing them unprotected is dangerous. Corporations and colonies maintain "hulls" — walled computational environments — and pay for immune-system services. Their protection depends on limited access, separate permissions, controlled entrances and continuing maintenance as well as defensive AI (see Protected networks, in Everyday technology). Each bloc patrols its own net, and the frontiers between them are as much digital as physical. Light-lag gives Proxima's net, the only one beyond the Sun, an ecology of its own, with its own endemic agents, like an island with unique species.

The Swarm Years are also why this civilisation fears self-replicating factories: agents that copied themselves on rented compute taught every bloc what an unsupervised copier does. When the Mirror Race needs replicators from the 2380s, every bloc builds them under human supervision at first, then cuts that supervision back as far as it dares, to keep pace with its rivals (see The Mirror Race).