| 标题 | Tencent AI-Infra-Guard 4.0 Information Disclosure (CWE-200) |
|---|
| 描述 | # Technical Details
A Sensitive Data Exposure vulnerability exists in the [GetTaskDetail](cci:1://file:///root/llm-project-python/AI-Infra-Guard/common/websocket/task_manager.go:1270:0-1357:1) method in [common/websocket/task_manager.go](cci:7://file:///root/llm-project-python/AI-Infra-Guard/common/websocket/task_manager.go:0:0-0:0) of AI-Infra-Guard.
The application fails to mask sensitive API tokens when returning task detail responses. While commit e5582e7 introduced `maskToken()` for the Model List API, the same protection was omitted for the Task Detail endpoint. When [GetTaskDetail](cci:1://file:///root/llm-project-python/AI-Infra-Guard/common/websocket/task_manager.go:1270:0-1357:1) is called, it unmarshals `session.Params` and returns it verbatim, including plaintext model API tokens.
# Vulnerable Code
File: common/websocket/task_manager.go
Method: GetTaskDetail
Why: The method unmarshals session.Params containing the raw API token and includes it in the HTTP response without any masking or redaction. The existing maskToken() function used in the Model List API was not applied here.
# Reproduction
1. Submit a task via POST /api/v1/app/taskapi/tasks with a known API token in the model configuration.
2. Retrieve the task detail via GET /api/v1/app/tasks/{sessionId}.
3. Observe the plaintext API token in the response JSON under params.model.token.
# Impact
- Credential Leak: AI model API keys are exposed in plaintext to any user who can access the task detail endpoint.
- Financial Loss: Stolen API keys can be used to consume paid API quota.
|
|---|
| 来源 | ⚠️ https://gist.github.com/YLChen-007/fe4b834144ad535d167507c2008d4011 |
|---|
| 用户 | Eric-y (UID 95889) |
|---|
| 提交 | 2026-03-20 15時48分 (7 月前) |
|---|
| 管理 | 2026-04-04 23時33分 (15 days later) |
|---|
| 状态 | 已接受 |
|---|
| VulDB条目 | 355384 [Tencent AI-Infra-Guard 4.0 Task Detail Endpoint task_manager.go 信息公开] |
|---|
| 积分 | 20 |
|---|