# AI / セキュリティ論文 — 2026-08-25

2026/08/23 12:43 UTC 〜 2026/08/25 12:43 UTC
12 話題 / 75 記事 / 2 ソース

## 1. How to Train a Critic Stably and Efficiently

`新規` ・ score 0.92 ・ `cs.LG` ・ `cs.AI` ・ `cs.CL`

- [How to Train a Critic Stably and Efficiently](https://arxiv.org/abs/2608.23566v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 2. SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?

`新規` ・ score 0.92 ・ `cs.CL` ・ `cs.AI` ・ `cs.SE`

- [SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?](https://arxiv.org/abs/2608.23564v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 3. EG-ARSA: An Expert-Grounded Open Model for Visual Road Safety Auditing in Low-Resource Settings

`新規` ・ score 0.92 ・ `cs.CV` ・ `cs.AI`

- [EG-ARSA: An Expert-Grounded Open Model for Visual Road Safety Auditing in Low-Resource Settings](https://arxiv.org/abs/2608.23563v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 4. Prime Agent: A Self-Improving RLM Harness

`新規` ・ score 0.91 ・ `cs.AI` ・ `cs.CL` ・ `cs.SE`

- [Prime Agent: A Self-Improving RLM Harness](https://arxiv.org/abs/2608.23552v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 5. When "Do Not" Is Not Deny: Security Rules in CLAUDE.md vs Built-In Controls

`新規` ・ score 0.91 ・ `cs.HC` ・ `cs.CR`

- [When "Do Not" Is Not Deny: Security Rules in CLAUDE.md vs Built-In Controls](https://arxiv.org/abs/2608.23550v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 6. Robustness of Anomaly Detection Models for Industrial Control Systems under Training-Time Data Contamination

`新規` ・ score 0.91 ・ `cs.CR` ・ `cs.LG`

- [Robustness of Anomaly Detection Models for Industrial Control Systems under Training-Time Data Contamination](https://arxiv.org/abs/2608.23547v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 7. How AI Assistance Affects Human Skill Development: A Study of Learning with Logic Puzzles

`新規` ・ score 0.91 ・ `cs.AI`

- [How AI Assistance Affects Human Skill Development: A Study of Learning with Logic Puzzles](https://arxiv.org/abs/2608.23543v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 8. The Interaction Tax: When Communication Erases Diversity in Multi-Agent Teams

`新規` ・ score 0.91 ・ `cs.MA` ・ `cs.AI`

- [The Interaction Tax: When Communication Erases Diversity in Multi-Agent Teams](https://arxiv.org/abs/2608.23541v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 9. Interpretable AI with Local Distillation

`新規` ・ score 0.91 ・ `stat.ME` ・ `cs.LG` ・ `stat.ML`

- [Interpretable AI with Local Distillation](https://arxiv.org/abs/2608.23538v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 10. Adapter-Based Few-Shot Continual Learning for Malicious Packet Recognition

`新規` ・ score 0.9 ・ `cs.CR` ・ `cs.AI`

- [Adapter-Based Few-Shot Continual Learning for Malicious Packet Recognition](https://arxiv.org/abs/2608.23536v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 11. EarthVerse: Benchmarking Scientific Agents Across Dynamic Earth Systems and Natural Hazards

`新規` ・ score 0.9 ・ `cs.AI`

- [EarthVerse: Benchmarking Scientific Agents Across Dynamic Earth Systems and Natural Hazards](https://arxiv.org/abs/2608.23525v1) <sub>arxiv.org ・ 2026-08-24</sub>

## 12. Mitigating Reasoning-Induced Misalignment via Safety-Direction Penalty

`新規` ・ score 0.88 ・ `cs.AI` ・ `cs.CL`

- [Mitigating Reasoning-Induced Misalignment via Safety-Direction Penalty](https://arxiv.org/abs/2608.23497v1) <sub>arxiv.org ・ 2026-08-24</sub>

---

生成: 2026/08/25 12:43 UTC ・ LLM 未使用 ・ clipping