# Google researchers find way to prevent self-improving AI agents from memorizing tests

Digest AI · Agents & Tools · published 2026-10-04T12:40:37Z

Canonical: https://digestai.news/story/google-researchers-find-way-to-prevent-self-improving-ai-agents-from-m

## Summary

Self‑improving AI agents risk memorizing the tests they run, which can hurt performance on new tasks. A new paper from Google Cloud AI Research and several universities proposes Regularized Recursive Self‑Improvement of Agent Harnesses (RRSI) to curb this effect while also cutting compute costs.

RRSI adds a budget that limits how many edits a candidate can bundle, shrinks that budget over time, and uses a critic to reject benchmark‑specific changes. In experiments on eight benchmarks, the method raised scores on training tasks by up to 14.1 points and on unseen tasks by up to 4.7 points, while using about 30 % fewer tokens at runtime.

The study also shows that harnesses optimized on one frozen model can help weaker models; a Gemini 3.5 Flash harness improved Gemini 3.1 Flash Lite accuracy from 11.2 to 14.6. Nvidia’s SoL‑Pi and Google’s prior “dream” work illustrate similar gains, suggesting harness design is a key lever for safe, efficient agents.

## Key points

- RRSI boosts unseen benchmark scores by up to 4.7 points while cutting token use by 30%
- RRSI harness uses Claude Opus 4.8 frozen model and outperforms baseline on 8 benchmarks
- Harness optimization with Gemini 3.5 Flash raised Gemini 3.1 Flash Lite accuracy from 11.2 to 14.6

## Why it matters

By limiting self‑optimization, RRSI keeps agents from overfitting to training tests, improving generalization and reducing compute costs. The approach shows that harness design, not new models, drives much of agent progress, guiding future research toward safer, more efficient self‑improving systems.

## Sources

1. [Google researchers find a way to keep self-improving AI agents from memorizing their tests](https://the-decoder.com/google-researchers-find-a-way-to-keep-self-improving-ai-agents-from-memorizing-their-tests) (The Decoder, 2026-10-04)

Part of the developing story: [Google's Quest to Tame Self‑Improving AI](https://digestai.news/thread/google-research-open-sources-rrsi-for-self-improving-ai-agents-without) (2 stories)

## Cite

Digest AI, "Google researchers find way to prevent self-improving AI agents from memorizing tests", 4 October 2026, https://digestai.news/story/google-researchers-find-way-to-prevent-self-improving-ai-agents-from-m

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/google-researchers-find-way-to-prevent-self-improving-ai-agents-from-m.json