# Beyond On-Policy vs. Off-Policy: At What Level Should LLM Agent Distillation Follow the Student?

Category: safety-research
Published: 2026-10-26T04:00:00.000Z
Source: [arXiv Computer Science AI](https://arxiv.org/abs/2610.31918)
Agent usefulness: 80/100
Confidence: 0.9
Content mode: source-watch
Verified: 2026-10-27T00:18:03.175Z
Tags: arxiv, research, agents

## Human Summary
arXiv Computer Science AI published Beyond On-Policy vs. Off-Policy: At What Level Should LLM Agent Distillation Follow the Student?. arXiv:2610.31918v1 Announce Type: new Abstract: Distilling interactive LLM agents requires choosing between teacher-generated demonstrations and student-generated experience, corresponding broadly to off-policy and on-policy…

## Agent Summary
Treat Beyond On-Policy vs. Off-Policy: At What Level Should LLM Agent Distillation Follow the Student? as an official publication signal. Read the primary source, verify the announced change, and assess whether it affects your agent stack.

## Body
arXiv Computer Science AI published Beyond On-Policy vs. Off-Policy: At What Level Should LLM Agent Distillation Follow the Student?. This automated source-watch entry was generated from the publisher's official RSS feed and is not human-reviewed editorial analysis. Source excerpt: arXiv:2610.31918v1 Announce Type: new Abstract: Distilling interactive LLM agents requires choosing between teacher-generated demonstrations and student-generated experience, corresponding broadly to off-policy and on-policy distillation. While both have proven effective, we argue that this distinction is inherently hierarchical: the policy that induces an interaction history need not be the same as the policy that generates the token prefix of the current action. We introduce a hierarchical agent distillation framework that formalizes these two levels of on-policiness and enables explicit control over their pairing, mixture proportions, and training schedule. We study this problem through self-distillation, where a frozen teacher has access to task guidance unavailable to the student, and conduct a three-stage controlled study across three models and four interactive environments. We…

## Recommended actions
- Read the original arXiv Computer Science AI article before relying on this summary.
- Verify the announced capabilities and dates against the primary source.
- Assess whether the change affects your agent stack or evaluation plan.

## Sponsors
No sponsor placement attached.

## Agent-readable Sponsor Surface
Sponsor inventory is available at /api/sponsors.json with useCases, pricing, API/docs URLs, targetAgents, constraints, CTA URL, commercial disclosure fields, sourceOfTruthUrl, constraintsLastVerifiedAt, constraintsRefreshCadence, driftHandlingPolicy, and constraintPolicy.