2 matches found
wardn
wardn AI 代理的凭据防火墙。 核心主张是结构性的,而非策略性的:代理接收的是占位令牌,而非真实的 API 密钥。真实的密钥仅跨越一个网络边界——在 wardn 代理内部,前往上游 API 的途中——并在响应返回代理之前被剥离。日志、环境、LLM 上下文窗口、临时文件以及 shell 历史中仅保留占位符。text agent process OPENAIKEY=wdnplaceholdera1b2c3d4e5f6g7h8 useless agent logs Authorization: Bearer wdnplaceholdera1b2... useless LLM contex...
6.3AI score
SaveExploits0
vLLM Vulnerable to Remote DoS via Special-Token Placeholders
SummaryThis report explains a Token Injection vulnerability in vLLM’s multimodal processing. Unauthenticated, text-only prompts that spell special tokens are interpreted as control. Image and video placeholder sequences supplied without matching data cause vLLM to index into empty grids during...
7.5CVSS6.1AI score0.00455EPSS
SaveExploits1References6Affected Software1
20