jiazhizhong/higress
1
0
Fork 0
mirror of https://github.com/alibaba/higress.git synced 2026-05-24 12:47:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat(ai-security-guard): enhance risk action resolution and support sensitive data masking (#3690)

Browse Source 
Co-authored-by: rinfx <yucheng.lxr@alibaba-inc.com>
This commit is contained in:
JianweiWang
2026-04-15 11:14:56 +08:00
committed by GitHub
parent e2beb6cd45
commit b1187cc14d
13 changed files with 5019 additions and 214 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
plugins/wasm-go/extensions/ai-security-guard/README_EN.md
View File

@@ -35,12 +35,38 @@ Plugin Priority: `300`
| `contentModerationLevelBar` | string | optional | max | contentModeration risk level threshold, `max`, `high`, `medium` or `low` |
| `promptAttackLevelBar` | string | optional | max | promptAttack risk level threshold， `max`, `high`, `medium` or `low` |
| `sensitiveDataLevelBar` | string | optional | S4 | sensitiveData risk level threshold, `S4`, `S3`, `S2` or `S1` |
| `customLabelLevelBar` | string | optional | max | Custom label detection risk level threshold, value can be max, high, medium, or low |
| `riskAction` | string | optional | block | Risk action, value can be `block` or `mask`. `block` means blocking requests based on risk level thresholds, `mask` means replacing sensitive fields with desensitized content when API returns mask suggestion. Note: masking only works with MultiModalGuard mode |
| `timeout` | int | optional | 2000 | timeout for lvwang service |
| `bufferLimit` | int | optional | 1000 | Limit the length of each text when calling the lvwang service |
| `consumerRequestCheckService` | map | optional | - | Specify specific request detection services for different consumers |
| `consumerResponseCheckService` | map | optional | - | Specify specific response detection services for different consumers |
| `consumerRiskLevel` | map | optional | - | Specify interception risk levels for different consumers in different dimensions |
Risk level explanations for each detection dimension:
- For content moderation and prompt attack detection (contentModeration, promptAttack):
- `max`: Detect request/response content but do not block
- `high`: Block when risk level is `high`
- `medium`: Block when risk level >= `medium`
- `low`: Block when risk level >= `low`
- For sensitive data detection (sensitiveData):
- `S4`: Detect request/response content but do not block
- `S3`: Block when risk level is `S3`
- `S2`: Block when risk level >= `S2`
- `S1`: Block when risk level >= `S1`
- For custom label detection (customLabel):
- `max`: Detect request/response content but do not block
- `high`: Block when custom label detection result risk level is `high`
- Note: The Alibaba Cloud API only returns `high` and `none` for the customLabel dimension, unlike other dimensions which have four levels. Set to `high` to block on detection hit, set to `max` to not block. `medium` and `low` are kept for configuration compatibility but will not be returned by the API.
- For risk action (riskAction):
- `block`: Block requests based on risk level thresholds for each dimension
- `mask`: Replace sensitive fields with desensitized content when API returns `Suggestion=mask`, still block when `Suggestion=block`
- Note: Masking only works with MultiModalGuard mode (action configured as MultiModalGuard), other modes do not support masking
### Deny Response Body
When content is blocked, the plugin (`MultiModalGuard` action) returns the following structured JSON object. The location in the response depends on the protocol: