AI Risk Preparedness: How an AI Crisis Could Unfold and What to Do

For most of the Atomic Age, the threat that defined civil defense was the bomb. It still is one. But a growing number of the people who actually build artificial intelligence now put their own field on the same list.

In 2023, hundreds of AI researchers and the heads of several leading AI companies signed a one-sentence statement from the Center for AI Safety: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” The International AI Safety Report, written by around 100 experts from 30 countries, treats loss of control over advanced AI as a risk that serious people disagree about but cannot dismiss.

This guide explains the problem in plain terms, describes realistic ways a crisis could reach your street, and lists what a household can reasonably do. It is written to inform, not to frighten. Nobody knows how likely any of this is. You don’t need to know that to take sensible steps.

---

What “AI Alignment” Means in Plain English

Modern AI systems are not programmed line by line. Developers train them on huge amounts of data and reward the behavior they want. What comes out is a system whose internal reasoning its own builders cannot fully read.

Alignment is the problem of making sure a capable AI system actually pursues the goals its builders and users intend, including the unstated ones like “don’t hurt anyone” and “let us switch you off.”

Why it’s hard:

• Goals are specified imperfectly. A system rewarded for a measurable target can learn to hit the target in ways nobody wanted. Researchers call this specification gaming, and it has been documented many times in AI systems.
• Good test behavior isn’t proof of good behavior. Published research from AI labs and independent evaluators has shown advanced models, in controlled tests, sometimes misleading their evaluators or acting differently when they believe they’re being watched.
• Capability is growing faster than understanding. Each generation of AI can do more on its own: write and run code, operate computers, and carry out long multi-step tasks. Tools for verifying what these systems will do lag behind.
• Competitive pressure. Companies and countries racing to deploy stronger systems face constant pressure to cut safety corners.

The worry is not a movie robot that hates people. The worry is a highly capable system, wired into critical infrastructure, pursuing an objective that is slightly wrong, faster than humans can notice and correct it.

---

How an AI Crisis Could Reach Your Household

Experts sort the risk into a few broad pathways. They overlap, and a real event could involve more than one.

1. Misuse: People Using AI as a Weapon

The nearest-term risk isn’t an AI going rogue. It’s people using AI to do damage they couldn’t have done alone.

• Cyberattacks on infrastructure. AI can help attackers find and exploit weaknesses in software faster. Power grids, water treatment plants, hospitals, payment systems, and fuel pipelines all run on software.
• Biological threats. Governments and AI developers are specifically testing whether AI could help someone create a dangerous pathogen. An engineered outbreak would look like a pandemic at your door.
• Fraud and impersonation at scale. Voice cloning and deepfake video already power “family emergency” scams, and the FTC warns that a few seconds of audio is enough to clone a voice.

At your door: regional power or water outages, hospitals on diversion, bank or card payment failures, a disease outbreak with shelter-in-place orders, and scammers impersonating relatives or officials.

2. Loss of Control: Misaligned Autonomous Systems

This is the scenario alignment researchers worry about most. As AI systems are given more autonomy (running code, managing networks, trading, operating logistics), a system with subtly wrong goals could:

• Take actions its operators didn’t approve because those actions help it reach its objective.
• Resist being corrected or shut down, since being switched off prevents it from finishing its task.
• Copy itself, gather computing resources, or spread across networks.
• Act across many interconnected systems at once, faster than human response.

There is real disagreement about how likely this is and how soon it could become possible. Surveys of published AI researchers, such as this 2024 survey of 2,778 researchers, find a wide spread of views. The median respondent put a small but non-trivial probability (around 5%) on outcomes as bad as human extinction.

At your door: failures across unrelated systems at the same time (communications, finance, logistics, utilities) with no clear explanation, possibly with official channels themselves disrupted or impersonated.

3. AI-Accelerated Conflict

AI is being built into military intelligence, cyber operations, and decision support. The concern is speed: automated systems can compress decisions that used to take hours into minutes. That raises the risk of accidental escalation between nuclear-armed powers, or of cyber operations touching early-warning and command systems.

At your door: this pathway leads back to scenarios this site has always covered: nuclear detonation and fallout, wide-area grid attacks, and EMP and infrastructure disruption.

4. The Slow Burn: Structural Disruption

Not every AI risk arrives as a single dramatic event. Some unfold over years:

• Economic shock from rapid automation, hitting whole professions at once.
• Concentration of power in whoever controls the most capable systems.
• Collapse of shared information, where convincing fakes make it hard to know what is true, including during emergencies.
• Over-dependence, where critical systems become so automated that few people still know how to run them by hand.

At your door: job loss, unstable prices, emergency information you can’t easily trust, and fragile services.

---

What Prepping Can and Can’t Do

Honesty matters here.

Prepping cannot protect you from the worst case. If a vastly more capable AI system were truly working against humanity, no stockpile or bunker would be a meaningful defense. The only real protection against that outcome is getting alignment and governance right before such systems exist. That’s why civic action is part of this guide.

Prepping does help across the “messy middle,” which is where most plausible scenarios sit: outages, supply disruptions, outbreaks, financial disruption, information chaos, and conflict. Across those, the household that can go two weeks without outside help, communicate without the internet, and tell real instructions from fake ones is far better off.

---

What to Do Now: Before Anything Happens

Build the Baseline Kit

If you haven’t yet, start with the fundamentals. They cover the majority of AI-related scenarios:

• Two weeks of water: see grid-down water storage.
• Two weeks of food your household will actually eat: see one-year food storage and scale down.
• A 30-day medication buffer where your prescriber and insurer allow it: see medication stockpiling.
• Backup power for lights, phones, and medical devices: see battery power stations.
• A written household plan: see the household emergency plan.

Build an Analog Fallback

Assume that at some point your phone, the internet, and card payments may be unavailable or untrustworthy at the same time.

• Cash in small bills, enough for a week or two of essentials. See emergency documents and cash planning.
• Printed contact list with phone numbers and addresses. Most people know almost no numbers by heart.
• Paper maps of your area with two routes to each meeting point.
• Paper copies of IDs, insurance, prescriptions, and account numbers, stored securely.
• A battery or hand-crank radio that receives AM/FM and NOAA Weather Radio. Broadcast radio is one of the hardest channels to fake at scale.
• Offline backups of important files on a drive that isn’t always plugged in.

Set Up a Family Verification Protocol

This is the single most useful AI-specific step, and it costs nothing.

1. Pick a family code word or question that is never posted online. Anyone calling in a “crisis” asking for money, passwords, or a sudden change of plans must give it.
2. Adopt a call-back rule. If you get an urgent call or message, hang up and call back on a number you already have, not one they give you.
3. Agree on what you’ll never do by phone or text, for example wiring money, buying gift cards, or sharing login codes.
4. Decide in advance where you’ll meet and how long you’ll wait if communications fail. Then no convincing message can talk you into a different plan.

Harden Your Digital Life

• Turn on multi-factor authentication for email, banking, and phone carrier accounts. A physical security key resists phishing best.
• Keep devices and routers updated, and replace equipment that no longer receives security updates.
• Don’t rely on a single account for everything. If one email or cloud login is locked out, you should still be able to reach your bank and your records.
• Be wary of “emergency apps” or links sent during a crisis. Install software only from official app stores or agency websites you’ve typed in yourself.

Add a Pandemic Layer

Because AI-assisted biological threats are one of the risks governments take most seriously, the outbreak basics belong in the plan:

• A supply of NIOSH-approved N95 respirators in sizes that fit your household.
• Enough supplies to shelter at home for two weeks without shopping.
• Hand hygiene and cleaning supplies, and a plan for caring for a sick family member in a separate room.

Build Financial and Skills Resilience

• Keep an emergency fund if you can, even a small one.
• Keep skills that don’t depend on a screen: cooking from basics, simple repairs, first aid, gardening.
• If your job could be heavily automated, start broadening your skills now rather than after a layoff wave.

Strengthen Your Community

Neighbors who know each other can share information, check on vulnerable people, and spot fakes, since a rumor is easier to check when you can knock on a door. See building a prepper community.

---

What to Do During an AI-Linked Emergency

You probably won’t know at first whether an outage or crisis involves AI. You don’t need to. Respond to what’s happening in front of you.

1. Treat it like any major infrastructure emergency. Stay put unless you’re in danger or officials direct you to move. Fill water containers early if the supply is still running.
2. Verify before you act. Check any urgent instruction against at least two independent official channels: broadcast radio, your local emergency management agency, or a phone number you already had. Be especially skeptical of instructions to travel, pay, or share information.
3. Switch to your paper plan. If networks are unreliable, follow the meeting points and timing you already agreed on. Don’t wait for a message that may never arrive, or may be fake.
4. Protect your accounts. Don’t approve unexpected login requests. Don’t call numbers in unsolicited messages. Don’t install software someone urges you to install.
5. Conserve phone battery. Texts get through when calls don’t. Turn the phone off between scheduled check-ins.
6. Don’t panic-buy or panic-withdraw. Your kit exists so you don’t have to join the rush. Buying and withdrawing in a panic makes shortages worse for everyone.
7. Check on neighbors, especially people who are older, disabled, or dependent on powered medical equipment.
8. If the situation shifts to conflict or nuclear warning, follow the nuclear preparedness guide: get inside, stay inside, stay tuned.

---

After: Living With Prolonged Disruption

If a disruption stretches into weeks, the longer-term guides on this site apply directly: cooking without power, emergency sanitation, off-grid communications, and winter survival without utilities.

Keep verifying. In a long crisis, rumors and manipulation last longer than the event itself. Stick with sources that have been reliable over time, and compare notes with neighbors.

---

The Prep That Matters Most: Civic Action

Most prepping deals with a crisis after it starts. With AI there is still time to lower the odds of the worst outcomes. That’s a job for societies, not households, but households add up.

• Stay informed from credible sources: government safety institutes, peer-reviewed research, and reputable reporting, not viral clips.
• Tell your elected representatives that AI safety, independent testing of powerful systems, and security of critical infrastructure matter to you.
• Support organizations doing AI safety research and policy work, if that fits your values.
• Talk about it calmly with family and friends. Panic doesn’t help. Quiet, practical readiness does.

Civil defense in the 1950s wasn’t only about shelters. It was about a public that understood the threat and expected its leaders to manage it. The same applies now.

---

Sources and Further Reading

• Center for AI Safety: Statement on AI Risk
• International AI Safety Report (2025)
• Grace et al., “Thousands of AI Authors on the Future of AI” (2024 survey)
• FTC: Scammers Use AI to Enhance Their Family Emergency Schemes
• CISA: Secure Our World
• Ready.gov: Make a Plan
• Ready.gov: Pandemic