Build Research
Account
Live AI risk

Automated probes, outage data and public reports, reviewed by a person.

Open the monitor
AI risk nowLevel 1 · Calm

25 systems, 0 regions, 0 reports in 24 h.
Before it fails

What to do in the first 72 hours if AI-dependent services fail.

Read the guides
Offline kitSave it now

Guides and data to keep when the internet is down.

Research

Why Day Zero exists: preparing for an uncontrolled AI crisis

Toswo watches AI risk. Day Zero is the plan for the worst case: an AI-related crisis that gets out of control. What it is, what it is not.

Short answer

Toswo is an artificial intelligence centre that monitors AI risk. Day Zero is the part of the site for the worst case that monitoring cannot prevent: an AI-related crisis that gets out of control and takes away the systems people rely on, such as power, networks, payments, logistics and information. It is a plan for that day, written calmly.

We do not know if or when this will happen. Most AI incidents recorded so far are small and local. Day Zero exists the way a fire exit exists: you hope never to use it, and it is much easier to find before the alarm.

Why an AI centre has this section

An organisation that tells people when AI systems fail should also tell them what to do when failure is not small. That is the only reason this section exists. It is not a general survival site. For floods, storms, earthquakes and ordinary outages, use your government's emergency service and the Red Cross or Ready.gov first; many of our guides help there too, but they were not written for that.

What we mean by an "uncontrolled AI crisis"

No single event. It is a family of situations, in rough order of how plausible they are today:

  1. A large AI-assisted cyberattack on infrastructure, causing days or weeks of power or network outages.
  2. A cascading failure of AI-dependent services (cloud, payments, logistics), so shops, cash machines and deliveries stop for a while.
  3. A flood of false information that makes it hard to know what is true during an emergency.
  4. Autonomous or "agentic" systems acting outside their limits, with sudden economic or operational shocks.
  5. The worst case, loss of control of very advanced AI. This is hypothetical. No household can solve it by preparing; what protects people is that others work on AI safety, and what Day Zero can do is keep knowledge, people and small groups able to recover.

What each one would mean for you, and what helps.

What is known and what is not

  • Many AI researchers and company leaders signed the one-sentence Statement on AI Risk in 2023: that reducing the risk of extinction from AI should be a global priority alongside pandemics and nuclear war. Others think the concern is overstated. We do not claim to settle that.
  • The International AI Safety Report treats loss of control as a hypothetical scenario, and at the time of its 2026 edition reported that fully autonomous, end-to-end cyberattacks had not been seen, while criminals and state groups already use AI to help with attacks.
  • So the honest position is: uncertain, low-probability for the severe cases, worth cheap preparation.

What Day Zero is not

  • Not a prediction and not a reason to panic. Nothing here says a crisis is near.
  • Not a replacement for emergency services. Where they work, call them first.
  • Not medical advice. See how these guides are reviewed.

What you get

  1. An offline copy of knowledge on a drive you can read without the internet: medicine, water, food, tools, mathematics, science, maps. See Downloads.
  2. A timeline so you always know the next job: the first 72 hours, 7 days, 30 days, 6 months, one year.
  3. A shopping list to build before it happens: what to store.
  4. A way to work with other people: groups and teamwork and finding other groups.
  5. Honest limits: what an offline archive can rebuild, and what it cannot.
  6. A rehearsal: the weekend drill.
  7. Answers in plain language: the Day Zero questions, and how this compares with other resources.

How it connects to the monitor

The AI risk monitor shows today's level and every event behind it. Day Zero is for what to do if events escalate past what monitoring can contain. Most days you will only need the monitor. If you only do one thing, download the kit once while the internet works, and check it again each month.

The order of priorities

Every plan in these guides follows what kills people fastest:

  1. Injury and exposure (hours)
  2. Water (days)
  3. Sanitation and disease (days to weeks)
  4. Food (weeks)
  5. Shelter and fuel (weeks, sooner in winter)
  6. People: organisation, trust, rest (always)
  7. Tools, trade and knowledge (months to years)

When you do not know what to do next, ask which of these is least secure and fix that.