Genko 原稿Back to edition

REALM: Identifying Minimal Task Assets for LLM Agents

Maps each task's required assets, cutting prompt-injection success from 16.6 percent to 0.2 percent.

Explainer · on the wire

The explainer for this story is still on the wire.

The claim, the mechanism, and the figures land here once generated.

Source · arXiv

arxiv.org/abs/2610.39155cs.CR · cs.AI · cs.SY · eess.SY

More from this edition

REALM: Identifying Minimal Task Assets for LLM Agents — Genko 1 Nov 2026