← All papers  /  Nov 17, 2026
Memory · Agents

GUARD: Stabilizing Recursive Self-Improvement with Regret Guidance

First page
GUARD: Stabilizing Recursive Self-Improvement with Regret Guidance
The curator’s take

Minghe Zhang, Xuetao Kang, Chen-Yu Lee and colleagues at Google Cloud present GUARD, a grader-free loop that edits an agent harness (prompts, skills, tool rules, memory) from experience and only accepts an edit when its gains outweigh the regressions it causes.

Ask this paper

Key points
01

Regret instead of scalar scores. Each task is split into pre-execution self-checks. Every round grades trajectories against those checks, pins rules the agent has mastered and ranks newly broken behaviors first, so an edit that fixes one case and breaks another is visible as a regression.

02

Soft-HURT gate. Parallel candidate edits are compared by group-relative fitness, and the gate promotes an edit only when gains exceed regressions on its buffer. Pooling two candidates in parallel cut rejected gate rounds by 64% against sequential search at equal budget.

03

Memory without pollution. On procedural memory tasks GUARD leads task accuracy (83.0%) and update correctness (90.1%) with 0.000 procedural pollution. Aggressive rewriters such as Meta-Harness reach similar update scores but triple pollution.

04

Hosted SDK result. On 43 test tasks GUARD reaches 80.80% accuracy, 8.89 points over the starting harness (paired p = 0.008) and 6.28 points over Meta-Harness, while using 54.5% fewer tokens per task. At test time one frozen harness runs in a single pass.

Abstract

Autonomous AI agents pair a frozen base model with an editable harness of prompts, skills, tool rules, and memory. Recursive self-improvement (RSI) adapts this harness from experience without weight updates, yet self-editing without external graders often backfires: reflection and scalar scores conflate what an edit newly broke with what remains unsolved, causing local regressions, skill forgetting, and memory bloat. We introduce GUARD (Grader-free Unsupervised Environment Design [UED] with Adaptive Rule Distillation), a closed-loop framework for reference-free continual harness evolution. Our key insight is to decompose each task into pre-execution self-checks and guide harness updates via directional regret. Each round grades trajectories against these checks, pinning mastered rules while prioritizing newly broken behaviors, and selects among parallel candidate edits by critic-free group-relative fitness. A Soft-HURT gate then promotes an edit only when gains outweigh regressions, scoping specialized fixes to prevent over-triggering and guaranteeing non-regression on its gated buffer. At test time, one frozen harness runs in a single pass. Across benchmarks on memory, skill evolution, continual task streams, and domain reasoning, GUARD consistently improves performance without catastrophic forgetting or memory pollution while substantially cutting deployment token usage.

Every Monday
Get next week’s papers.
Subscribe on Substack