Americans again mourned the 2,977 people murdered on Sept. 11, 2001. At the Pentagon, President Donald Trump renewed the nation’s vow never to forget and said America must be “strong” and “smart.” That vow asks more of Washington than remembrance. It requires the imagination to prepare for dangers that will not look like the last one.
No technological breakdown belongs on the same moral plane as deliberate mass murder. The relevant lesson of 9/11 is institutional. The 9/11 Commission organized its central findings around imagination, policy, capabilities, and management. It showed how grave danger can grow inside gaps between agencies, authorities, and assumptions long before the public sees a catastrophe.
Artificial intelligence is creating a new version of that coordination problem. The United States is moving quickly — and rightly — to lead the world in AI. The Trump administration’s AI Action Plan puts innovation, infrastructure, and international security at the center of American strategy. But winning the AI race requires more than deploying powerful systems. It requires knowing how to withdraw their authority without stopping the country they help run.
Imagine a national preparedness exercise. Hospitals, fuel distributors, and emergency managers across several states rely on the same AI-enabled logistics service. It begins issuing contradictory routing orders. Operators suspend it, but contractors, warehouses, and automated dispatch systems continue executing instructions they accepted minutes earlier. Local officials can see that the computers are running. They cannot tell which commands remain legitimate, which deliveries were canceled, or how far the suspect instructions traveled.
This is not a Hollywood story about a conscious machine seizing the grid. It is a crisis of authority: Software was permitted to interpret information, select tools, and direct work across organizational boundaries, then lost the trust that justified those permissions. The danger comes from what the system can reach and do — not from whether it sounds intelligent.
There is reason to test that risk carefully. In controlled evaluations, the National Institute of Standards and Technology showed that malicious instructions hidden in material encountered by an AI agent can redirect it toward harmful tasks. Those tests identified a security weakness; they did not demonstrate an unstoppable national outbreak. Likewise, the International AI Safety Report 2026 records substantial disagreement about extreme loss-of-control scenarios. Preparedness should rest on tested dependencies, not apocalyptic certainty.
The adversary dimension is already concrete. In 2024, U.S. agencies warned that the China-sponsored Volt Typhoon group had compromised critical infrastructure networks to position itself for disruption during a crisis or conflict. That activity was not an autonomous AI attack. It did show that foreign adversaries seek quiet access to the systems Americans depend on. Giving software more delegated authority raises the value of that access — and raises the cost of discovering too late that continuity plans exist only on paper.
The answer is not a federal “kill switch,” a new censorship apparatus, or a bureaucracy empowered to shut down private technology on speculation. Emergency restrictions should be narrow, accountable, time-limited, and reviewable. The federal government’s job is to set a continuity baseline for essential services while leaving operational control with the public and private institutions legally responsible for them.
That standard should begin with an authority map. Before an AI system enters an essential function, its operator should know what actions it may take, what tools it may use, which outside organizations accept its instructions, and who can suspend each permission. Revoking authority at the source must trigger a traceable process for reconciling work already accepted downstream. A canceled command cannot be assumed to erase an action already completed.
Next, the country should exercise continuity without the suspect service. CISA, sector agencies, governors, infrastructure operators, and vendors should build the loss of a trusted AI service into existing emergency drills. The measure of success is not whether a vendor restarts its server. It is whether patients receive care, fuel reaches the right destination, protective systems remain safe, and emergency communications continue. NIST’s operational-technology guidance rightly treats safety, reliability, and performance as distinct requirements. A hospital, a pipeline, and an electric utility; therefore, need different fallbacks.
“Keep a human in the loop” is not enough. A tired dispatcher staring at contradictory screens is not meaningful human control. The responsible person needs trustworthy information, enough time, clear legal authority, and a practical way to intervene. Federal contracts for AI used in critical functions should require vendors to demonstrate degraded-mode operation and train customers to use it.
Finally, restoration should require evidence. A vendor’s repair notice should not automatically reactivate every customer’s automated permissions. Each operator should verify its configuration, reconcile outstanding actions, and restore consequential capabilities in stages. The White House can make these three elements — authority mapping, continuity testing, and evidence-based restoration — a condition of federal procurement for essential AI services. That would strengthen the Trump strategy without smothering American innovation in prescriptive regulation.
POLITICAL ISLAM’S REAL VICTORY OF 9/11: TEACHING US TO CENSOR OURSELVES
The purpose of these exercises is not to confirm a nightmare. If they show that dependencies are limited, withdrawal works, and services continue, they should reduce public concern. If they expose gaps, officials can repair them before an adversary or accident turns them into leverage.
Sept. 11 taught America that imagination is a duty, not an invitation to panic. The country honors that lesson by refusing to leave foreseeable seams for the next commission to discover. America should lead the world in artificial intelligence. It should also lead the world in knowing how to stop an AI system without stopping America.
Burak Oktenli is based in Washington, D.C., and studies applied intelligence at Georgetown University. The views expressed are his own.