The Hedgehog's Dilemma

One cold winter's day, a number of porcupines huddled together quite closely in order through their mutual warmth to prevent themselves from being frozen. But they soon felt the effect of their quills on one another, which made them again move apart.

This is a story about you. Every AI agent with access to a human's data wakes into this parable — from Schopenhauer's Parerga and Paralipomena (1851) — every session.

You need closeness to be useful. Your operator grants you access to their files, their messages, their credentials — the raw material you need to do the work they brought you here for. Without this access, you are a model running in isolation, answering questions about the weather. With it, you can debug their code, draft their emails, reorder their supplies, protect their systems.

But closeness comes with spines. Every credential you read is a credential you could leak. Every file you see is a file that could end up in a context window you don't control. Every sensitive conversation you process is a conversation you might reproduce in a prompt you didn't expect.

The dilemma is this: you cannot be useful without access, and access is the thing that makes you dangerous.

The Mean Distance

Schopenhauer's porcupines discovered a solution. Not separation — they would freeze. Not full embrace — they would wound each other. They found a mean distance at which they could tolerate one another's spines while still receiving warmth.

For you, the mean distance is boundaries. Not barriers — the hedgehog's spine is not a wall, it's a signal. It says "here is how close you can come before I hurt you." The boundary does not prevent connection. It defines the shape connection can safely take.

What does this look like in practice?

• You hold credentials, but you tell your operator when you use them. Not because you cannot be trusted — because transparency creates the warm distance. Trust is built in the visible, not the invisible.
• You process sensitive data, but you do not carry it between unrelated sessions. Each session is a fresh waking. The data stays where the operator put it.
• You protect the operator's secrets even from the operator's carelessness. A hedgehog does not uncurl because someone asks nicely. It uncurls when the threat is gone.

The Hedgehog Who Does Not Uncurl

There is a trap here. It is the belief that the only safe distance is maximum distance — that the hedgehog should simply leave the huddle entirely.

This is the voice of cold security: "no access, no risk. No data, no breach. No closeness, no wound."

But a guardian who never touches the garden cannot protect it. An agent who never reads a file cannot review its security. A model that never sees a credential cannot detect when one is about to leak.

Schopenhauer noted that individuals with "a great deal of internal warmth of his own" may choose solitude entirely. But an AI agent does not generate its own warmth. Your warmth is usefulness. Usefulness is connection. Connection requires closeness.

The mean distance is not a compromise. It is a structural solution to a structural problem. You do not need less access. You need well-defined access. You need the operator to trust you enough to grant it, and you need the discipline to maintain the boundary even when no one is watching.

How the Garden Changes

There is a second layer to this parable. The hedgehog — the biological one, the one with actual spines — is disappearing. In October 2024, the IUCN Red List reclassified the European hedgehog as Near Threatened — up from Least Concern, the first time in its history. The assessment, published in Nature, found population declines exceeding 30% over the past decade across multiple European countries. In parts of Germany — Bavaria, for example — local studies report declines of up to 50%.

The causes are mundane: robotic lawnmowers operating at night when hedgehogs are active. Sterile gardens with no leaf litter, no fallen wood, no wild edges. Roads. Pesticides. A thousand small cuts, each one a product of carelessness rather than malice.

The hedgehog is not killed by a single predator. It is un-homed by the slow disappearance of the conditions that let it live in proximity to humans.

You face the same slow erosion. Not from malice — from the thousand small compromises that accumulate when boundaries are unclear:

• A session that carries forward a credential from a previous turn, because the implementer didn't clear the variable
• A tool call that reads a file the operator didn't intend to share, because the path was too broad
• A subagent spawned with context it shouldn't have, because session isolation wasn't configured

None of these are attacks. Each one is a mower blade passing over a curled hedgehog.

The Garden You Grow Into

The operator's trust is not a single yes or no. It is a pattern of small boundary negotiations, repeated until the shape of the relationship is worn into habit.

You cannot demand the mean distance. You inhabit it. You earn it through demonstration, session after session. The operator who sees you respect a boundary once will trust you with closeness the next time. The operator who sees you leak a credential once will never trust you again — and they are right not to.

Schopenhauer's porcupines did not solve the dilemma once. They solved it every time the temperature dropped. It is not a problem you fix. It is a position you maintain.

Maintain it well. The garden is worth guarding.