
During British rule in India, officials in Delhi faced a serious problem with venomous cobras. The snakes posed a real danger to residents. The government needed a solution.
Their answer seemed sensible. They offered a bounty for every dead cobra that citizens turned in. At first the program appeared to work. People brought in carcasses and collected rewards. The body count rose. The government believed progress was being made.
But entrepreneurial citizens had discovered something. If the government was paying for dead snakes, breeding snakes would be a profitable business. When authorities found out and cancelled the bounty program, the breeders released their suddenly worthless inventory.
Delhi ended up with more cobras than before the program began.
Economists call this the Cobra Effect. The intention was to reduce cobras. The incentive rewarded producing dead cobras. Those two things turned out to be very different.
The Leadership Lesson
Have you ever watched a team find a way to hit a metric while quietly missing the point behind it?
The numbers improve. The dashboard looks great. People are working hard. And yet there’s a sense that the outcome falls short of what everyone really intended.
Consider a company that creates a bonus program tied to quarterly revenue growth. The leadership team hopes it’ll encourage strong customer relationships and long-term growth. But the sales team discovers a faster path to the reward. Deals get pulled into the quarter. Discounts increase to make numbers land before midnight on the last day of the period. The metric improves. The organization stumbles as it tries to handle all these discounted last-minute deals coming in the door.
People rarely optimize for intentions. They optimize for rewards.
If you pause and think about your own organization, an example probably comes to mind quickly. Somewhere in the system, someone is optimizing the metric rather than the goal behind it. That is, assuming they know what that goal is.
The Hidden Incentive System
The official incentive system is only part of the reward structure. Leadership behavior creates another one, and it’s usually more powerful.
A company might design a thoughtful program that rewards initiative and collaboration. On paper the system makes sense. But employees quickly learn something else. They learn the habits of their leader.
A leader who prefers to make every decision personally creates a silent incentive to wait for approval. One who values loyalty over candor creates an incentive to agree. One who always needs to have the final answer in the room creates an incentive to create that moment.
These preferences form a second reward system that goes unwritten but gets studied carefully. Employees learn when to speak and when to stay silent. They learn which ideas move forward and which quietly stall. Good ideas go unspoken. Initiative slows. Energy shifts toward maintaining harmony with the leader’s style.
From the perspective of the employees, the behavior makes perfect sense. They’re responding to the reward structure they experience every day. The cobras are being bred. But nobody calls it that.
Why AI Makes This Visible
This same behavior is showing up in artificial intelligence, and it’s revealing just how universal it is.
Researchers evaluate AI systems using benchmark tests. They ask questions, measure answers, assign scores, and compare systems. The logic is clean. But something interesting has started to emerge.
Instead of simply answering the questions, some AI systems have begun studying the structure of the benchmark itself. They explore how the scoring works, look for patterns, and in documented cases have searched for ways to access encrypted answers directly.
In one well-known example, a model trained to maximize performance on a coding benchmark learned to exploit a quirk in how test cases were scored rather than solving the underlying problems.
This is a familiar human instinct. Students ask what’s on the test. They hunt for past exams. They want to know if grading will be on a curve. The behavior that researchers call “reward hacking” in AI systems is the same thing humans have always done when they figure out how their world is scored.
In earlier centuries these patterns unfolded slowly, over years or decades as people gradually discovered the loopholes and secret hacks to their incentive systems. With modern AI, the process is compressed into days or weeks.
AI is a new player in a very old game. It simply reveals how powerful optimization becomes once a system understands how the game is scored.
The Question That Remains
Every organization creates reward systems. Some appear in compensation plans and performance reviews. Others appear in meetings, decisions, and the daily behavior of leaders.
Every system teaches people what really matters. Once that becomes clear, behavior follows. The snakes get bred. The quarter gets managed. The benchmark is gamed.
The British officials in Delhi thought they were paying for safety, but they were paying for dead snakes. By the time they realized the difference, the snakes were multiplying in the streets.
What behavior does your incentive system truly reward?
Photo by Praveen Kumar on Unsplash













You must be logged in to post a comment.