The Internet Almost Died in 1997

The Internet Runs on Trust

To understand how the internet almost died in 1997, you need to understand one thing: the internet has no central authority deciding where data goes. There is no master switchboard operator. Instead, tens of thousands of independent networks — called Autonomous Systems — cooperate by telling each other what they can reach.

The protocol they use for this conversation is called BGP, the Border Gateway Protocol. Created in 1989, BGP is sometimes described as the "postal service of the internet." Each network announces: "I can deliver mail to these addresses." Neighbouring networks note this and pass it along: "Our neighbour says they can reach those addresses, so we can too."

There is no verification. No authentication. No cryptographic proof that a network actually can reach the destinations it claims. If a router announces that it's the best path to Google, other routers will believe it and start sending Google-bound traffic that way.

April 25, 1997: The Day the Internet Believed a Lie

MAI Network Services operated as AS7007 — just one of thousands of Autonomous Systems on the internet. On that April morning, a technician made a configuration error that caused their router to do something catastrophic: it absorbed the entire global routing table — every known route on the internet — and re-announced all of them as if MAI was the origin.

In BGP terms, MAI's router told the world: "I am the shortest path to every single destination on the internet."

The internet believed it.

Traffic from across North America and Europe began funnelling toward MAI's network in Florida. Their routers, designed to handle the traffic of a mid-tier ISP, were suddenly expected to process a significant fraction of all internet traffic. They collapsed almost instantly.

But the cascading failure was worse than a simple overload. Because routers everywhere had updated their tables to point through MAI, even after MAI's network went dark, the corrupted routes persisted. Traffic aimed at legitimate destinations was being sent into a black hole. Routers began "flapping" — rapidly switching between the bad routes and attempts to find alternatives — which consumed processing power and caused even more routers to fail.

The Fix (And Why It Took So Long)

The engineers who noticed the problem couldn't simply "undo" the bad routes from a central location. There is no central location. Each of the thousands of affected networks had to independently recognise the bad routes and filter them out.

In practice, this meant humans at ISPs around the world had to manually intervene — identifying MAI's bogus announcements, filtering them, and waiting for the corrected routing tables to propagate. The process took roughly two hours for most of the internet to recover, though some networks experienced disruptions for much longer.

It Keeps Happening

The AS7007 incident was a wake-up call. But decades later, BGP remains largely trust-based. Notable repeat incidents include:

• 2008: Pakistan Telecom tried to block YouTube domestically by announcing a more specific route for YouTube's IP space. The announcement leaked to the global internet, redirecting YouTube traffic worldwide into a black hole. YouTube was down globally for about two hours.
• 2017: A Russian AS accidentally hijacked routes for Google, Apple, Facebook, and Microsoft, rerouting their traffic through Russia for several minutes.
• 2018: A Nigerian ISP's misconfiguration caused Google traffic to be routed through China and Russia for over an hour.

Solutions exist — a framework called RPKI (Resource Public Key Infrastructure) allows networks to cryptographically verify route announcements. But adoption has been slow. As of the mid-2020s, a substantial portion of the internet's routes remain unprotected by RPKI.

The internet's greatest vulnerability isn't sophisticated hackers. It's a technician having a bad morning.