Forward-looking agents anticipate a change and pre-adjust their behaviour to offset its intended effect, so the net effect is engineered-minus-offset. The behavioural response set is part of the system being acted upon, so any analysis holding behaviour fixed systematically overstates the intervention's efficacy.
How would you explain it like I'm…
Hide It First
Anticipatory neutralization is when people see a change coming and quietly get ready for it ahead of time, so the change ends up doing nothing. Imagine a teacher says 'tomorrow I'll take away one cookie from anyone holding cookies' — so today everyone eats their cookies early. When tomorrow comes, there's nothing left to take, and the rule does nothing. The people didn't fight the rule; they just got ahead of it.
Adjust Early, Cancel The Change
Anticipatory neutralization is when people can see a rule or change coming, and they adjust their behavior in advance to cancel out what it was supposed to do. The real effect on the world is the planned effect MINUS however much people pre-adjusted to dodge it. If they fully dodge it, the change does nothing at all, even though it was carried out correctly; if they partly dodge it, you get a smaller effect than expected; sometimes they over-react and it even backfires. The catch is that people's reactions are part of the system you're trying to change, so any prediction that pretends people won't react will guess too high. It doesn't take anyone being sneaky — just enough warning to adjust, a way to adjust, and a reason to.
Pre-Adjusting To Offset A Policy
Anticipatory neutralization is the pattern where forward-looking agents anticipate a coming change and pre-adjust their behavior to cancel out its intended effect. The net effect equals the engineered effect minus the agents' anticipatory offset: a complete offset makes a correctly-implemented intervention show no effect; a partial offset leaves some effect; an over-compensation can even flip the sign. The key idea is that people's responses are part of the system you're acting on — they re-optimize under the new expected regime before the change lands — so any analysis that holds behavior fixed will systematically overstate the intervention's effect, because the offset is selected precisely to reduce it. It doesn't require anyone to be sneaky or sophisticated; it needs only three things: enough lead time to adjust, an adjustment lever the agent controls, and adjusting being in the agent's own interest.
Anticipatory neutralization is the pattern in which forward-looking agents anticipate a policy or environmental change and pre-adjust their behaviour so as to fully or partially neutralise the change's intended effect. The intervention's net effect on outcomes is the engineered effect minus the agents' anticipatory offset. When the offset is complete, the intervention has no observable effect even though it has been correctly implemented; when partial, it delivers the engineered effect minus a behavioural compensation; when over-compensating, it can even reverse direction. The structural commitment is that the behavioural response set is itself part of the system being acted upon: a forward-looking agent re-optimises under the new anticipated regime before the intervention takes effect, so any analysis that holds behaviour fixed will systematically mis-predict the outcome — and the error will be in the direction of overstating the intervention's effect, because the offset is structurally selected to reduce that effect. The mechanism does not require the agent to be malicious, dishonest, or technically sophisticated. It requires only three conditions: that the agent anticipates the change with enough lead time to adjust, that the agent has an adjustment lever within its control budget, and that adjusting serves the agent's existing interests better than letting the change land unmodified. When those three conditions hold, the offset arises mechanically rather than through coordination. This is what makes the pattern substrate-portable: the lever and the interests vary across domains, but the conditions under which pre-adjustment neutralises an engineered effect are the same.
Separates four things policy discourse collapses: the intended effect (under fixed behaviour), the engineered effect, the behavioural response, and the net observed effect — forcing the analyst to compute the offset rather than assume it away.
Collapses a family of policy disappointments into one diagnostic — which agents anticipated this, what levers do they hold, how does adjustment serve their interests? — and sorts the repair catalogue: reduce anticipation, constrain the lever, or realign the interest.
Predicts which interventions land cleanly (those touching parameters agents cannot adjust) and fixes the direction of mis-prediction: anticipatory offset always reduces effect relative to fixed-behaviour models.
Technology/A-B testing: the Lucas-critique question "do agents re-optimise when the policy changes?" exposes announcement-driven bias in long-running tests.
Cyber-defence: the Peltzman discipline predicts users granted two-factor authentication will relax password hygiene.
Organisational change: the monetary surprise-versus-expected distinction explains why pre-announced changes generate the very pre-emptive behaviour they must fight.
A deficit-financed tax cut is engineered to raise demand, but households anticipating that today's debt means future taxes save the windfall to smooth consumption — so the stimulus is partially or fully neutralised, and a fixed-behaviour model overstates it.
Anticipatory Neutralization is not Antifragility because antifragility is a system gaining capacity from stressors, whereas anticipatory neutralization subtracts an intervention's efficacy by pre-adjustment.
Anticipatory Neutralization is not Goodhart gaming because gaming corrupts a proxy metric, whereas anticipatory neutralization is legitimate forward-looking re-optimisation of actual behaviour.
Anticipatory Neutralization is not Moral Hazard because moral hazard is the insurance-specific instance, whereas this prime is the general forward-looking-offset parent of which moral hazard is one case.