The non-security popular press, exemplified by The Atlantic, has begun highlighting the darker side of AI beyond friendly chatbots. AI tools excel at writing code and thus at executing cyberattacks, leading to a fourfold increase in daily attacks identified by Palo Alto Networks from 2024 to 2025. Hackers now deploy adaptive AI-enhanced viruses, automate espionage, and steal data far faster than before.
Experts like Alex Stamos note a crazy amount of offensive activity, with companies hacked daily. This stems from decades of slapdash software engineering where vulnerabilities persist for years because skilled hackers were scarce. AI upends this balance by making discovery and exploitation rapid, reducing time-to-exploit from over 700 days in 2020 to 44 days in 2025.
Governments and firms are alarmed by advanced models like Claude Mythos Preview and GPT-5.5-Cyber, which match elite human hackers and remain restricted. Open-source tools already empower low-skill criminals. Incidents like the Canvas hack by ShinyHunters and Meta AI giving access to 30,000 Instagram accounts illustrate the risks. The next months to years will likely bring more severe outages, especially for under-resourced entities like hospitals and utilities.
Individuals can mitigate with password managers, updates, and thin clients, but collective action at Y2K scale is needed urgently. No one knows if good or bad actors will find remaining bugs first, or if new AI capabilities will continually uncover more complex flaws.