The Cobra Effect · Horst Siebert

The cobra effect: when the fix makes the problem worse

Curated by · reviewed 2026-06-01

A reward aimed at solving a problem can backfire by rewarding the wrong behavior. Pay people to kill cobras and someone starts breeding cobras. Whenever you incentivize a proxy, expect people to game the proxy — sometimes worse than the original problem.

The cobra effect: a well-intentioned incentive backfires because people optimize the literal reward rather than the outcome you wanted — sometimes making the problem worse than before you intervened.

The story (told by economist Horst Siebert): colonial Delhi had too many venomous cobras, so the government offered a bounty for dead cobras. It worked briefly — then enterprising locals began breeding cobras to cash in. When officials discovered the scheme and scrapped the bounty, the breeders released their now-worthless snakes, leaving more cobras than at the start. The fix didn't just fail; it reversed. A real version happened in Hanoi with rats, where a bounty paid for severed rat tails produced tailless rats roaming the city — breeding more rats to harvest.

It's a vivid case of perverse incentives, close kin to Goodhart's law: the moment you reward a measurable proxy ('dead cobras') instead of the true goal ('fewer cobras'), people optimize the proxy by whatever means, including means that defeat your purpose. Pay developers per bug fixed and you may get bugs introduced to be fixed. Reward call centers for closed tickets and problems get closed, not solved. Fund agencies by the size of the problem they manage and the problem mysteriously never shrinks. People aren't being evil — they're responding rationally to exactly what you rewarded.

The defense is to design incentives by asking 'how would a clever, self-interested person game this?' before launching it — run the cobra test. Reward outcomes you actually want, not easy proxies; watch for the behavior the incentive will breed, not just the behavior you hope for; and stay ready to kill an incentive that's being gamed (carefully — even removing it can backfire, as the released cobras showed). Whenever you attach a reward to a number, assume the number, not the goal, is what you'll get.

Why it matters

Every incentive you set — at work, in policy, in your own life — risks rewarding the wrong thing; the cobra effect is the warning to design for how people will actually game the reward, not for how you hope they'll behave.

A common misreading

It's not 'all incentives backfire, so never use them.' Well-designed incentives work powerfully. The lesson is to design for the behavior the reward will actually breed — anticipate gaming, reward true outcomes over proxies — not to abandon incentives, which would just leave the original problem unaddressed.

Put it to work

Test yourself

What is the cobra effect?
Show answer
When an incentive meant to fix a problem makes it worse — because people optimize the literal reward (dead cobras) rather than the real goal (fewer cobras), even by means that defeat the purpose.

Reading it once isn’t remembering it.

Lock this idea into memory with a 5-minute active-recall session — the science of spaced repetition, no signup.

Try this idea free →
Worth remembering? Post on X
Embed this idea on your site

A self-contained card that links back here — paste it anywhere:

Related ideas

Goodhart's law: when a measure becomes a target, it stops measuringThe Tyranny of Metrics Second-order thinking: always ask 'and then what?'The Most Important Thing What everyone owns, no one protectsThe Tragedy of the Commons

FAQ

What is the cobra effect?
A perverse incentive where a reward meant to solve a problem backfires. The classic tale: a bounty on cobras led people to breed cobras for the reward, leaving more cobras than before.
How is the cobra effect related to Goodhart's law?
Both describe what happens when you reward a proxy instead of the true goal. Goodhart: a measure that becomes a target stops measuring. The cobra effect is the vivid case where gaming the proxy makes the original problem worse.
Was this helpful?