# Google Research Open-Sources RRSI: AI Agents That Improve Their Own Harness Without Overfitting

> Source: <https://www.marktechpost.com/2026/09/29/google-research-open-sources-rrsi-ai-agents-that-improve-their-own-harness-without-overfitting/>
> Published: 2026-09-29 09:01:36+00:00

**Google Cloud AI Research**, with UNC-Chapel Hill, Stanford and Washington University in St. Louis, has released [RRSI (Regularized Recursive Self-Improvement)](https://regularized-rsi.com/). It lets an LLM agent rewrite its own harness: prompts, tools, memory, control flow and sub-agents. Model weights never change. RRSI constrains the improvement loop itself, so gains hold on benchmarks the agent never optimized against.

**Deployable?** Yes, as a research framework. The [code is Apache 2.0](https://github.com/google-research/rrsi), needs Python 3.10+, and accepts any [LiteLLM](https://github.com/BerriAI/litellm) model string. Defaults assume Claude Opus 4.8 on Vertex AI. 

## **Why Self-Improving Harnesses Overfit**

Harness evolution loops propose edits, score them on a fixed evolve set and keep the winner. The same tasks are reused every round, so the loop can memorize them. The [RRSI research](https://arxiv.org/abs/2609.24972) names 3 failure modes: benchmark-specific fitting, noise chasing and complexity accumulation. Each one widens the gap between evolve-set scores and real transfer.

## **How RRSI Works**

RRSI keeps every harness component editable. It regularizes how the search moves instead.

#### **Proposal side**

- **Annealed edit budget:** a cosine schedule lets early rounds bundle several edits. Late rounds allow a single attributable change.
- **Evidence-aware credit:** each candidate is logged with its component, hypothesis, diff, score change and cost change. The proposer reads this ledger, so falsified ideas are not retried.
- **Structured exploration:** when progress stalls inside the noise band, budget shifts to components the run never touched.

#### **Selection side**

- **Leakage critic:** rejects task names, entities, answers or benchmark-specific logic before any scoring.
- **Noise-adjusted floor:** gains must clear the variance measured on the unchanged base harness.
- **Cost rule:** extra inference tokens must be paid for by measured gain.
- **Pruning:** components that stop producing gains become deletion targets.

The research team frame these as analogies to classic regularizers. The edit budget maps to L0, pruning to Lasso (L1) and the cost rule to Ridge (L2).

## **Results Across 8 Benchmarks**

- [Terminal-Bench 2.1](https://github.com/harbor-framework/terminal-bench) (evolve split): 74.2% to 80.2%.
- [SWE-bench Verified](https://github.com/SWE-bench/SWE-bench) (never used for selection): 82.0% to 83.8%.
- Out of distribution: [JobBench](https://github.com/Job-Bench/job-bench-eval) +4.7,[GDPval](https://openai.com/index/gdpval/) +3.5 and[APEX-Agents](https://www.mercor.com/apex/apex-agents-leaderboard/) +3.7 points.
- [EngDesign](https://github.com/AGI4Engineering/EngDesign) (evolve) +4.9;[Frontier-Eng](https://github.com/Einsia/Frontier-Engineering) +4.3 Medal points.
- [Harvey LAB](https://github.com/harveyai/harvey-labs) : +1.1 on the evolve split, +2.3 on its held-out split.

All 6 held-out splits improved. With [Gemini 3.5 Flash](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/) as the policy, Terminal-Bench 2.1 rose from 64.6 to 78.7. SWE-bench Verified rose from 76.8 to 79.0.

The harness is also lighter. On the agentic workspace instance, RRSI uses 2.42M policy tokens per trial. Unregularized evolution uses 3.80M. The abstract reports this as 30% fewer; the project page says 36%.

## **RRSI vs Closest Competitors**

Scores come from Table 1 of the RRSI research paper. All methods share the same starting harness, policy, evolve split and candidate budget.

| Feature | RRSI | [Meta-Harness](https://arxiv.org/abs/2603.28052) | [AHE](https://arxiv.org/abs/2604.25850) | [TTHE](https://arxiv.org/abs/2607.08124) | [HarnessX](https://arxiv.org/abs/2606.14249) | 
|---|---|---|---|---|---|
| **Core idea** | Regularized proposal and selection | Agentic proposer over code, scores and traces of all prior candidates | Observability-driven loop; edits paired with verified predictions | Evolves harness during test time, no gold labels | Modular typed primitives, trace-driven adaptation | 
| **Model weights** | Frozen | Frozen | Frozen | Frozen | Frozen | 
| **Cost rule and pruning** | Yes | No* | No* | No* | No* | 
| **Harvey LAB evolve score** | 90.5 | 93.0 | 90.7 | 91.1 | 91.8 | 
| **OOD average (H0 = 39.7)** | **43.6** | 40.6 | 39.2 | 38.0 | 39.7 | 

**Per the RRSI research team. OOD average is the mean of JobBench, GDPval and APEX-Agents, computed from [Table 1](https://arxiv.org/html/2609.24972v2).*

Meta-Harness leads on the Harvey LAB evolve split. RRSI has the smallest evolve gain but the only OOD average more than 1 point above H0.

## **Interactive Explainer**

### How RRSI Regularizes Agent Self-Improvement

**Proposer** Reads full edit ledger, shrinking edit budget

**Leakage critic** Screens diff before scoring

**Evaluate** Run on evolve set

**Gate** Noise floor + cost rule

**Harness H** Accepted, then pruning check

<sub>t+1</sub>
<sub>t</sub>= ⌈ b

<sub>min</sub>+ (b

<sub>max</sub>− b

<sub>min</sub>) · ½(1 + cos(πt / T)) ⌉

<sub>0</sub>+ β

<sub>1</sub>ΔS

<sub>0</sub>fixed at 5% here). In RRSI, δ is estimated on the unchanged base harness and β values are tuned on the evolve set, then frozen.

[RRSI paper](https://arxiv.org/abs/2609.24972)·

[GitHub](https://github.com/google-research/rrsi)·

[project page](https://regularized-rsi.com/)Built by Marktechpost

## **Getting Started**

```
git clone https://github.com/google-research/rrsi.git && cd rrsi
pip install -e ".[dev]"
python3 rrsi.py --domain coding baseline
python3 rrsi.py --domain coding run
```

Each round drafts 2 candidates in separate git worktrees, screens them, evaluates both and fast-forwards the branch to the winner. The coding instance also needs Docker and [harbor](https://github.com/laude-institute/harbor). New domains plug in through a single adapter module.

## **Key Takeaways**

- RRSI evolves prompts, tools, memory and workflows while model weights stay frozen.
- A leakage critic, noise floor, cost rule and pruning decide which edits stick.
- Terminal-Bench 2.1 rose from 74.2% to 80.2% with Claude Opus 4.8.
- SWE-bench Verified, never used for selection, rose from 82.0% to 83.8%.
- Apache 2.0 code on GitHub; research-grade, not an official Google product.

Check out the **[Paper](https://arxiv.org/pdf/2609.24972)**, **[Codes](https://github.com/google-research/rrsi)** and **[Technical details](https://regularized-rsi.com/).** All credit goes to the researcher of this project. Also, feel free to follow us on **[Twitter](https://x.com/intent/follow?screen_name=marktechpost)** and don’t forget to join our **[150k+ML SubReddit](https://www.reddit.com/r/machinelearningnews/)** and Subscribe to **[our Newsletter](https://magic.beehiiv.com/v1/f5e63dd4-5653-4f09-83e2-321a8b1ba526?email={{email}})**. Wait! are you on telegram? [now you can join us on telegram as well.](https://t.me/machinelearningresearchnews)

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? [Connect with us](https://forms.gle/MJjjVDPS7whH8Ngs6)

Asif Razzaq is the CEO of Marktechpost AI Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.