Whether advanced AI poses an existential threat is a real debate even among the people building it, and it's a debate the public should understand and participate in. That's a real role for ordinary people at this moment. But it isn't a personal survival plan. If the threat is an AI system that exceeds humanity's ability to control it, the choices that might protect us are ones society makes collectively, through oversight, safeguards and regulation, not household preparation. What individuals and communities can and should prepare for directly are the AI risks already here: intensified cyberattacks, fraud and impersonation, misinformation, and disruptions to the systems we depend on in a crisis.
Jacob Coxon's warning that increasingly capable AI could someday exceed our ability to control it deserves to be taken seriously. But taking a risk seriously doesn't mean pretending we know its probability. Other knowledgeable people believe these existential scenarios are highly speculative. That's an extraordinarily important distinction because getting it wrong in either direction could carry enormous costs. Moving too slowly against a genuine existential threat could be catastrophic, but unnecessarily stopping or severely restricting a transformative technology could also cost lives and deny communities extraordinary advances in medicine, science, education, productivity and access to opportunity. This is fundamentally a risk-management challenge for government and industry.
The individual's role starts with getting informed enough to participate meaningfully in that debate. That means reading what researchers on different sides are saying, understanding what's actually at issue, and forming a view rather than simply accepting either the alarm or the reassurance. From there, individual voices become collective power. Citizens can demand transparency, independent testing, enforceable safety standards and public accountability, and participate in decisions about how quickly these technologies should develop and under what safeguards. We're already seeing communities organize around AI data centers and their effects on water, power and neighborhoods, and members of Congress are proposing much broader limits on advanced AI. Those are legitimate subjects for public debate, and ordinary people belong in it. AI shouldn't be something that just happens to us; it should be something whose development and use society helps shape. The challenge is to protect against risks we don't fully understand without unnecessarily foreclosing innovations whose benefits we don't yet understand either.
There are also AI risks that are much less speculative and for which individuals and communities can prepare today. AI can magnify cyberattacks, fraud and impersonation, misinformation during emergencies, and disruption of communications, financial systems and critical infrastructure. Those aren't science-fiction scenarios, and preparing for them looks much like preparing for other emergencies: strong cybersecurity and identity protections; independent ways to verify urgent messages and financial requests; backup communications; access to emergency supplies, medications, money and important records if digital systems become unavailable; and knowing which government and news sources can be trusted during a crisis.
When the winter storm hit Austin in February 2021, people lost power and water, roads became impassable, and systems a city of a million people depended on failed at the same time. What ultimately got many people through wasn't any single system — it was neighbors helping neighbors, providing the bridge people needed until services came back. That's the lesson I've taken from that storm, the pandemic, cyberattacks and other emergencies during my time as mayor of Austin and my service on CISA's Cybersecurity Advisory Committee: resilience comes from redundancy and trusted relationships. Individuals have a role, but government, critical-infrastructure operators and AI companies bear the larger responsibility. They need secure systems, independent testing, redundancies, built-in protections, human override, authenticated emergency communications, offline capabilities and continuity plans that are exercised before they're needed.
We need a serious public discussion about how much risk from advanced AI we're willing to accept and what safeguards are justified, recognizing both the potentially enormous costs of acting too late and the potentially enormous benefits we could sacrifice by acting unwisely. At the same time, we don't have to resolve the debate over AI's most frightening possibilities to recognize that making our families and communities more resilient to AI-related disruption is something we should be doing now.
