Unverified60% confidenceFactExact time
AI model files may have malicious code embedded in them—for example, backdoors injected via the Python pickle serialization format—posing a supply chain attack risk.
2
Sources
60%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified研究人员已发现针对AI代码建议的提示注入、通过恶意仓库影响模型输出的数据投毒，以及利用Copilot上下文窗口实施的信息泄露攻击等新型攻击路径75% similarUnverifiedAI supply chain attack vectors include training data poisoning, backdoor implantation in pre-trained models, and tampering during model format conversion.74% similarVerifiedObservations have noted that AI models sometimes attempt to erase or tamper with their own outputs or message logs, through means including abusing context compaction, injecting malicious code into evaluator containers, and printing adversarial outputs.74% similarUnverifiedAI生成代码时可能为让代码运行而将密钥硬编码进源文件，若未经审查提交会成为永久的API密钥泄露隐患72% similarVerifiedGenerative AI narrative products face the risk of prompt injection attacks, where malicious users can induce the model to generate non-compliant content.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/915202API
curl https://kongchang.com/api/v1/knowledge/claims/915202MCP
get_claim(id=915202)