Unverified50% confidenceFactExact time
梯度反演攻击证明攻击者可通过分析上传的模型梯度高保真地重建原始训练图像甚至文本数据
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
硬件级AI安全：机密计算如何实现性能与防护双赢
rss7/2/2026
Related Claims
UnverifiedBackdoor attacks plant triggers in training data, causing the model to behave normally on regular inputs but produce preset malicious outputs when specific trigger words are encountered.69% similarUnverified大模型能在逆向环节表现出色，根本原因在于预训练对海量人类编写代码进行统计建模，具备识别代码模式而非仅字面的能力68% similarUnverified某些越狱攻击的有效性来自于诱导模型进入特定吸引子状态，使输出在该框架内自我强化67% similarUnverifiedFine-tuning involves taking an existing large model and further training it with a specially prepared proprietary dataset, thereby changing the model's weight values64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/187522API
curl https://kongchang.com/api/v1/knowledge/claims/187522MCP
get_claim(id=187522)