Will an AI catastrophe force a global internet shutdown before 2028?
Check out the FutureSearch Results
1%
Probability: Very Unlikely Weighed against ~300 sources and related frontier-AI forecasts. The answer, with high confidence, is no.
Why the answer is almost certainly no
A YES outcome requires an unprecedented conjunction of two events: an AI system causing catastrophic global damage, and authorities responding with a deliberate, multi-country disconnection of internet infrastructure attributed specifically to that AI. Each is individually improbable in the next 18 months. Together, they are vanishingly so.
Capabilities aren't there yet
AI offensive capabilities are advancing, but remain far below the threshold for systemic global collapse. The most significant AI-orchestrated cyber incident to date — the espionage campaign disrupted by Anthropic in late 2025 — still relied heavily on humans, hit a limited set of targets, and damaged no infrastructure.
Defenders contain, they don't disconnect
Standard incident-response playbooks emphasize targeted containment: isolating affected networks, air-gapping specific systems. A broad internet shutdown would be economically ruinous and technically complex — the opposite of established cyber-defense doctrine.
Kill switches are local, not global
Serious proposals for AI "kill switches" (e.g. RAND's analyses) focus on data-center-level disconnection, explicitly avoiding broad shutdowns. Bodies like the Internet Society actively push to maintain connectivity to limit cross-border fallout.
Recent outages don't count
The 2024 CrowdStrike outage and the 2025–26 Cloudflare disruptions were configuration errors, not malicious AI. And the hundreds of internet shutdowns each year are single-country political blackouts — censorship, not catastrophe response.
What the warnings actually tell us
The Five Eyes, the IMF, and the Bank of England have all raised legitimate anticipatory warnings about AI accelerating cyber exploitation and amplifying systemic financial risk. But warnings are not incidents. To date, no AI-driven event has materialized as a cascading global catastrophe — and the ~18-month horizon to January 2028 leaves very little room for both a discontinuous leap in AI capability and a complete reversal of global incident-response doctrine.
The bottom line
The 1% is not zero. It accounts for one tail scenario: a sudden, unexpectedly severe jump in AI capability triggering a panicked, overreactive international response. Everything we know about how defenders, regulators, and network operators actually behave argues against it.
The internet is far more likely to be defended by staying connected than by being switched off.
Forecast held at 1% · ~300 sources reviewed · Horizon: Jan 1, 2028
Gerd Leonhard · For the Human Future gerdleonhard.com · gerd.tv
Gerd says… 35%
