Will AI Kill Us?

How an AI warning changes as it passes through posts, interviews, headlines and retellings.

Distinguishing dates in forecasts and hypothetical scenarios from claims about what will happen.

A comparison of researchers, reports and risk estimates, with attention to uncertainty and time horizons.

The case for taking advanced AI risks seriously, including autonomy, reward hacking and failures of oversight.

Arguments against the most alarming scenarios, including limits of evidence, deployment constraints and disagreement among forecasters.

The author weighs the competing arguments and uses a claim-check table to separate evidence from interpretation.

The incidents discussed in the book and the lessons it draws about permissions, monitoring and security controls.

Examining employment claims, the difference between tasks and roles, and the value of experience and judgment.

Impersonation, voice cloning and practical verification habits for families and workplaces.

Questions to ask about data handling, the information to protect and why tool settings and service tiers matter.

Checking provenance and corroboration, and recognising instructions planted inside material an AI system reads.

Using AI for drafts, explanations and everyday administration while checking important output.

Turning notes into actions, adapting documents for an audience and keeping responsibility for the finished work.

Using AI to support learning, building family habits and encouraging children to explain what they understand.

Judgment, verification, accountability and relationships in work increasingly supported by AI.

Moving from producing more output to directing, reviewing and taking responsibility for useful work.

A four-week practice plan for testing a tool, finding its limits and improving one repeatable task.

Defining a business problem, a success measure and an acceptable risk before buying AI tools.

Assessing uses by their stakes and reversibility, with different levels of human oversight.

Common mistakes involving confidential data, untested deployments, unrestricted permissions and unchecked output.

A one-page policy covering approved tools, sensitive information, review responsibilities and incident handling.

The book examines policy approaches, competition among labs and countries, and arguments about slowing development.

Five possible futures presented as scenarios, with signs to watch and questions about preparation.

The closing argument for staying attentive, accepting uncertainty and taking specific practical steps.