A manager abdicates when a system’s score replaces the judgment owed to a worker. A school abdicates when detection software replaces the harder work of teaching students how to think. A newsroom abdicates when a plausible image is treated as evidence because it fits the mood of the hour. A government abdicates when private systems shape public life without public accountability. A military abdicates when lethal force is routed through automation until responsibility becomes too distributed to name.
Leo is especially direct about private power. He warns that the main drivers of technological development are private and often transnational actors with resources and capacities that can surpass those of many governments. That is not an argument against innovation. It is a description of public authority being transferred, piece by piece, into privately owned systems.⁶
A teacher does not meet that power as a theory. She meets it on a Tuesday night, reading essays that all sound smoother than the drafts her students used to write. One paper is honest. One is copied. One is partly generated and partly revised by a student who no longer knows where assistance ended and thought began. The software may flag a probability, but it cannot restore the missing conversation. She still has to ask the student what he meant, how he got there, and whether the sentence on the page belongs to his mind or only passed through his account.
That is the human problem inside the technical one. AI unsettles the relationship between evidence and trust. It changes what a teacher can assume, what a reporter can publish, what a manager can measure, what a judge can weigh, and what a citizen can believe.
The labor question follows from that. An AI system in an office may arrive as help, then become a quota, then become surveillance, then become evidence that fewer people are needed to do the same amount of work. The IMF estimated in 2024 that AI would affect almost 40 percent of jobs worldwide, replacing some work and complementing other work depending on policy, skill, and who captures the gains.⁷ The labor question is not whether every job disappears. It is whether the benefits are treated as a common good while the disruption is delivered as a private problem.
War is where abdication becomes most terrible. AI does not make violence morally cleaner because a model helped identify the target or an operator clicked after the system had already narrowed the choice. It can make force faster, more remote, more deniable, and easier to normalize. Leo warns of AI’s role in hybrid warfare, cyberattacks, information manipulation, influence campaigns, and automated strategic decisions.⁸ A weapon that sees only a target does not need hatred to kill. That is part of the danger.
This is why Leo’s call to “disarm” artificial intelligence carries more weight than the usual ethics language. “Disarm” can sound like an argument against technology. It is not. To disarm AI is not to pull the plug on every useful tool. It is to remove from the technology the presumption that speed outranks judgment, that efficiency outranks dignity, and that whatever can be automated should be automated.⁹
The best counterargument deserves more than a polite nod. AI may help rural clinics, disabled students, small newspapers, scientific researchers, and people who need tools that existing institutions have failed to provide. In some places, the alternative to machine assistance will not be perfect human care. It will be no care, late care, or care rationed by distance, money, and exhaustion. A country that refuses such tools in the name of purity would not become more humane. It would become weaker, slower, and less able to serve people who need help now.