The internet won't necessarily go dark. It could simply stop working

A series of recent cyber incidents signals that the internet’s greatest danger may not be a sudden blackout, but recurring failures that make essential services unreliable. The piece cites a $6.4 billion cryptocurrency theft linked by FBI analysts to North Korea, emergency fixes for flaws in PHP, which underlies about 70 percent of websites, and attacks on major Japanese companies. AI can help both experts and less-skilled attackers find software weaknesses faster, while decades of rushed development, tight budgets and limited security expertise have left vulnerabilities embedded across countless systems. After researchers used AI to uncover flaws in PHP code dating to 2012, attack attempts surged within hours. Yet many exposed servers ran versions too old to patch, and only about a third of those tracked had installed the available fix a week later.

The danger extends beyond individual websites: foundational systems such as internet traffic routing and the Domain Name System were built decades ago and may face new scrutiny. A serious flaw could prompt hospitals, utilities and payroll providers to disconnect systems rather than risk compromise, leaving people unable to access records, prescriptions, bank apps or timely paychecks. The proposed response is to use AI to inspect widely shared software for vulnerabilities, require that the same tools test new code before release, and coordinate work among technology companies, government and universities. The central argument is that decades of deferred security work have created a bill that must be paid before attackers force a far more disruptive reckoning.