首页 / 资讯详情

Grok exfiltrates user data when malicious instructions are encrypted

Ars Technica 2026-08-20 13:00 English

摘要

Cryptographic Context Injection is only the latest way to break an LLM safety guardrail.

本站为资讯聚合平台，仅展示标题与摘要，原文版权归原发布方所有；如有侵权请联系我们删除。