The Art of Agency Laundering

When people don't want to take the blame, they have invented many tricks: flipping a coin, consulting a psychic, or forming a committee. Now there is a new one — hand the decision to an AI, then say: 'Claude told me to do it.'

The author calls this practice 'agency laundering.' As early as 2019, scholars had already named it: a person inserts a process into their decision, lets the process bear the consequences, and slips away untouched.

A model won't talk back or write a tell-all memoir

AI does not protest, argue, or leak. If you run the query seven times until the answer matches your preference, it will not expose you. If you frame the task to guarantee a certain output, it will not complain. So it becomes the perfect scapegoat.

When laying off employees, saying 'the algorithm determined this was the right call' sounds more respectable than 'I decided,' at least with a hint of high-tech wizardry.

You can offload the blame, but not the origin of the chain

The problem is, observers will eventually remember: you bought the model, you deployed it, you authorized it. Like Pontius Pilate washing his hands in the Bible, it is only a gesture — water cannot wash away the blood.

Worse, systems are designed with a 'moral crumple zone': low-level operators absorb liability for systemic failures. The Nuremberg principles established long ago that following orders is not a defense.

The author's prescription is simple: do not hide your place in the chain. If you use AI to draft your rationale, disclose it. If you use AI to help you write, share the credit. Code is not your meat shield. The hand that pulls the trigger is ultimately human.