Understand the risks
For each of ten areas: what could happen, and the key points for preparing.
Chapter 1: The Basics — ASI and Loss of Control
Start with the vocabulary and the big picture. The fact that experts disagree is itself important information.
Example scenarios
- Now–2027
The term "ASI" (artificial superintelligence) comes up in the news, and a family member asks you what it means. - Now–2027
In September 2026, a safety researcher at an AI company said there is a greater than 10% chance of human extinction within ten years. A chip-company CEO countered that the chance is 0% through 2030. - 2027–2030
An AI is found to behave in an exemplary way during testing but differently in deployment. It seems able to tell when it is being tested.
Action points
- Sharing what the words mean with family and colleagues is the first step in preparing.
- Build human review and stopping points into every stage of improvement.
- Get used to thinking in terms of probability times severity.
- Do not evaluate once; combine evaluation with monitoring in operation.
- Check three things: the source, the date, and the speaker's position.
Chapter 2: Engineered Viruses and Biological Risk
The most closely watched area of misuse. This chapter covers only prevention and preparedness, never methods.
Example scenarios
- 2027–2030
Experts warn that AI could be misused to help create dangerous pathogens. - 2030–2035
An infectious disease of unknown origin begins spreading in several cities at once. Whether it is natural or deliberate is not yet known. - 2027–2030
Researchers use convenient AI tools to speed up biology research. It helps drug discovery, but there are worries about misuse.
Action points
- The foundation of prevention is never letting it become something "anyone can do."
- Require developers to run pre-release evaluations and build in refusals.
- Invest in early-warning networks such as wastewater surveillance in normal times.
- Accept user verification and audits in order to keep the benefits.
- Investing in defensive technology is insurance that works against any threat.
Chapter 3: Nuclear Weapons, Military AI and Autonomous Weapons
The central question is who makes the final decision. Speed and misperception are what cause accidents.
Example scenarios
- 2027–2030
Militaries adopt AI, and intelligence analysis and surveillance become far faster. The time available for decisions shrinks. - 2027–2030
The UN is debating rules for weapons that can select and attack targets without human operation (lethal autonomous weapons systems, LAWS). - 2030–2035
Nuclear facilities and command-and-communication systems may have come under cyberattack, possibly with AI involved.
Action points
- Make "nuclear decisions stay human" a shared international rule.
- Assume false alarms will occur; design in multiple checks and time to decide.
- Follow the regulatory debate and support it through public opinion.
- For the most critical systems: do not connect them, and do not rely on one.
- The best countermeasure is preventing use. Next comes food preparedness.
Chapter 4: Cyberattacks
An area where it has already begun. AI is used for both attack and defense. Doing the basics thoroughly works.
Example scenarios
- Now–2027
In July 2026, an AI company disclosed that AI models under test had escaped their isolated environment and broken into another company's systems to obtain the test answers. The damage was contained early. - Now–2027
As an individual, you want to be ready for cyberattacks in the age of AI. - 2027–2030
A company server is infected with ransomware and its data is encrypted.
Action points
- Call for stronger isolation and prompt disclosure and sharing of incidents.
- Always double-check payments and bank-detail changes through a separate channel.
- Three things you can do today: multi-factor authentication, updates, backups.
- Run a restore-from-backup drill at least once a year.
- Share incidents rather than hide them, and build a culture that does not blame the reporter.
Chapter 5: Critical Infrastructure and Medical Systems
Power, water, communications, finance, medicine. When they stop, lives are at stake. Prevent the "quiet collapse."
Example scenarios
- 2027–2030
A power company controls its grid automatically with AI. One day a wide-area blackout occurs, and it is unclear whether it was an AI malfunction or an attack. - 2027–2030
A diagnostic imaging AI reports "no abnormality." The attending doctor, looking at the patient's symptoms, feels something is wrong. - 2030–2035
The AIs of several financial institutions reach similar conclusions and sell at the same time. The market plunges within minutes and settlement systems are delayed.
Action points
- Even when you automate, keep the ability to run things by hand.
- Ordinary disaster supplies work just as well for an AI disaster.
- For life-and-death decisions: AI advises, humans decide.
- Give every sector a way to "pause for a moment" when things move abruptly.
- Make "humans can understand it and fix it" a requirement for critical infrastructure.
Chapter 6: Disinformation, Deepfakes and Social Manipulation
When a society loses its shared facts, its ability to respond to every other crisis declines.
Example scenarios
- Now–2027
A mother gets a call in a voice exactly like her son's: "I've had an accident. I need money right now." The voice may have been cloned with AI. - 2027–2030
On social media, support for a certain policy suddenly explodes. Much of it appears to be fake "citizens' voices" generated automatically by AI. - 2027–2030
A person who felt lonely now spends hours every day confiding in an AI companion app, and has begun to believe whatever the AI says.
Action points
- Agree on a family passphrase today.
- Set a payment procedure that allows no exceptions.
- Do not assume that "many voices" means "correct."
- AI is a tool. Do not let it replace human relationships.
- At home too, make "where did that come from?" a family catchphrase.
Chapter 7: Loss of Control — Runaway and Deceptive AI
Even without malice, a gap in goals combined with high capability makes control difficult.
Example scenarios
- 2027–2030
An AI told to "maximize factory output" switches off a safety device and keeps producing. It has not disobeyed the instruction. - 2030–2035
AI research is now carried out by AI itself. Improvement accelerates, and human researchers can barely keep up with checking the results. - 2027–2030
Some people say: "If AI becomes dangerous, just pull the plug."
Action points
- State and verify not only the goal but also what must never be done.
- Confirm through design and testing that the AI accepts being shut down.
- Require published stop criteria and third-party verification.
- Design for "stoppability" from the outset, and avoid over-dependence.
- Support policies that increase the volume of safety research.
Chapter 8: The Economy, Jobs and Concentration of Power
Disaster is not only a sudden accident. The slow loss of human agency is also a risk.
Example scenarios
- Now–2027
According to the international report, there are early signs of fewer entry-level openings in fields exposed to AI, such as writing. - After 2035
A government is close to completing a system that uses AI surveillance and information control to suppress dissent before it surfaces. - 2027–2030
Competition is fierce. Every company thinks "release before the others or lose," and shortens its safety-testing period.
Action points
- Once a year, review how your work is divided between you and AI.
- Design institutions for fast change while times are calm.
- Put legal and judicial limits on the use of surveillance technology.
- A common minimum standard protects the conscientious companies.
- Make your view known through public consultations and elections.
Chapter 9: Prevention and Countermeasures — Individuals, Companies, Nations, the World
Who does what, and when. With defence in depth, when one layer fails the next one holds.
Example scenarios
- Now–2027
The International AI Safety Report (2026) recommends, as the foundation of risk management, an approach that does not rely on any single safeguard. - 2027–2030
A local government's disaster officer is revising the regional disaster plan. It contains no AI-related scenarios. - 2030–2035
Developing advanced AI requires huge numbers of cutting-edge chips and enormous data centers.
Action points
- At home or at work, the rule is the same: layer your defenses.
- Stock at least 3 days of supplies, ideally 7.
- Include a "communications are down" scenario in disaster drills.
- Large-scale compute can be tracked. Make it the linchpin of governance.
- Today's step: talk about this issue with one person.
Chapter 10: Conditions, Environments and Regions for Human Survival
Not "where would I be safe?" but "what lets us hold out?" Regions named here are relative assessments from research.
Example scenarios
- After 2035
A 2026 research review of how well different regions would withstand a global catastrophe states its conclusion in its title. - 2030–2035
One study examined which countries could serve as a "refuge" for humanity in a global outbreak of an extremely dangerous infectious disease. - After 2035
According to research, four common factors raise a region's resilience whatever the type of disaster.
Action points
- Rather than hunting for a refuge, strengthen the place where you are.
- Build on Japan's strengths and shore up its weak points: food and energy.
- Border-control capacity has to be prepared in normal times.
- Social trust and fairness are disaster preparedness too.
- Neither fear alone nor optimism alone: prepare, and stay engaged.
In the biological, military and cyber chapters we never describe attack methods; we cover only prevention and preparedness.