Unverified50% confidenceTradeoffExact time

通过LoRA等参数高效微调方法，恶意行为体仅需少量计算资源和数据即可移除开源模型内置的安全对齐，使得在模型层面防止滥用对开源生态几乎完全失效

1

Sources

50%

Confidence

Long-term

Relevance

8/27/2026

First Seen

Sources

OpenAI Exposes Russian AI-Powered Influence Operations: Operational Patterns and Governance Challenges

Related Entities

Related Claims

Unverified下游用户对模型的二次微调可能意外破坏原有安全特性70% similar Unverified在决策覆盖攻击测试中，前沿模型表现更差，而该开源模型0%被覆盖决策70% similar Unverified工具调用使攻击面从让模型说错话升级为让模型做错事，危害从信息误导上升为实质性的数据访问和系统篡改68% similar Unverified从边际风险角度看，如果某项危险信息通过搜索引擎就能找到，模型是否开源对整体安全态势的影响并不显著68% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/810395

API

curl https://kongchang.com/api/v1/knowledge/claims/810395

MCP

get_claim(id=810395)