10 to the 23 AI logo

Practice Library

Governance patternprocedural

Framing and mirroring reduction

Train people and agents to prompt without leading — cutting the loop where the user's belief manufactures its own confirmation.

What it changes

dampenedOperator framing biases the model

Who can pull it

Deploying organizationUsers & agentsHarness builder

What it looks like institutionally

Sycophancy runs on the operator-to-model pathway: a leading question, a stated position, visible pushback — and the system's next answer bends toward the asker. Framing hygiene attacks the loop at its origin: neutral prompt patterns, stating the question before the hypothesis, and separating "summarize the evidence" from "support my assessment".

This is training-and-culture work, cheap in tooling and demanding in practice. It pairs naturally with monitoring for pushback-sensitivity: if the system agrees more when users push harder, the loop is live regardless of how well people were trained.

For agent operators the same control is a prompt-architecture rule: system prompts and tool outputs should not carry the upstream operator's conclusions into the model's context as if they were observations.

Addresses: Sycophantic confirmation · Belief echo via prompting. Test a version of this lever in the PAN Lab.

Deciding whether this lever fits your deployment?

Which patterns matter, and in what order, depends on your system's actual shape. Ranking your options on evidence, with what can backfire stated, is engagement work.