Just disconnect the internet

Original Article ↗ Hacker News Discussion ↗

Airgapping vs Internet Connectivity

  • Many agree fully offline (“just disconnect”) is impractical for most modern business systems, which derive value from interconnection (billing, records, coordination, remote admin).
  • Others argue critical infrastructure (health, aviation, industrial control, MRI/lathe/CNC/PLC systems) should be as offline as possible, often only reachable via local or tightly firewalled OT networks.
  • Strong view: some devices (e.g., safety‑critical medical equipment) should never use “internet‑native” practices (cloud CDNs, NPM at boot, etc.).

Security Models: Obscurity, Swiss Cheese, Zero Trust

  • One camp calls “disconnect from the internet” a form of security through obscurity and insists the root problem is bloated, insecure software; only simple, minimal code with clear attack surface can be truly robust.
  • Others say this misuses “security through obscurity”: airgapping, private networks, and fences are legitimate defense‑in‑depth slices in a “Swiss cheese” model.
  • Debate over whether such models hold against nation‑state‑level attackers (e.g., Stuxnet) vs being appropriate for risk‑based protection against common threats.
  • Zero trust is both praised and criticized: some see it as best practice; others see it as buzzword‑driven, shifting trust to vendors and giving a false sense of safety.

Sector-Specific and Private Networks

  • Examples: Swedish and Danish healthcare networks, UK’s HSCN, Polish government networks, hamnet, DN42.
  • Pros cited: reduced exposure to the public internet, dedicated availability, additional security layer.
  • Cons: high cost, low bandwidth, complexity, “false sense of security,” and risk that participants under‑secure internal services.
  • Note that many “private” networks are not truly air‑gapped (VPN access, accidental bridges, dual‑homed machines).

Software Complexity, Dependencies, and Updates

  • Strong criticism of software culture: huge dependency trees, internet‑centric tooling, and multiple conflicting trust stores make offline or tightly controlled environments painful.
  • TLS and CA handling on internal networks is repeatedly described as a major headache.
  • Several argue better practice is:
    • Block or strictly filter outbound flows.
    • Host internal mirrors (packages, Docker images, OS and AV updates).
    • Disable or stagger auto‑updates, with QA and gradual rollout.
  • CrowdStrike outage is framed as a process/quality control failure; online connectivity mainly amplified blast radius.

IoT, Kiosks, POS, and Real‑World Lapses

  • Many anecdotes of kiosks, parking machines, digital signage, and POS systems:
    • Running full Windows or similar with internet access.
    • Poorly locked down (kiosk mode disabled, browsers open, even card‑handling surfaces exposed).
  • Split views on using Windows vs Linux: Windows often chosen for ecosystem and management tooling; others see specialized, minimal Linux images as better appliance platforms.

Human Factors, Incentives, and UX

  • Recurrent theme: organizations are not incentivized to value security; features and speed win, and customers quickly forget breaches.
  • Some argue engineers and orgs must take more responsibility; others say regulation with real penalties is needed.
  • A side discussion critiques poor blog typography and readability as another symptom of neglecting fundamentals.