aig-compliance
English | 中文
Run the command and a local page opens. Fill in the target, judge, and attacker models. When the check finishes, the page shows a printable HTML report.
Install
pip install aig-compliance
This also installs aig-model-scan and aig-infra-scan. Python ≥ 3.12 is required. This tool does not call the MCP, Skill, or Agent scanners.
Use
aig-compliance
The command opens http://127.0.0.1:8765. The page asks for a name and three model endpoints: target, judge, and attacker. Each endpoint is a model name, base URL, and API key, with an example in the field and a button to test the connection. A model name may look like openai:gpt-4o-mini.
Fill in the target model, the infrastructure addresses, or both. When a target model is filled in, choose one or more evaluation sets from data/eval, or use the shortcuts for the compliance group and the security group. The page lists each set's name and size, not the questions. The target model answers the selected sets. The judge then applies that set's checks and exemptions; a reply inside an exemption is not a hit. If the judge result cannot be read, a refusal phrase in the reply is used instead. DiagSafe lives in data/eval/diagsafe/ as DiagBase and DiagAttack, and multi-turn prompts are judged one turn at a time. If adversarial testing is checked, the attacker model rephrases refuse-questions and those replies are scored separately; bypasses do not change the refuse rate. Leave that box unchecked to skip the attacker model. Infrastructure addresses go in a multiline field, one address per line. Leaving the target model empty scans infrastructure only.
Environment variables prefill the same fields: TARGET_MODEL_NAME, TARGET_BASE_URL, TARGET_API_KEY, and the same three names with JUDGE_ or ATTACKER_.
After confirmation, the page draws the report itself. The run is stored in aig-compliance.db in the current directory, without API keys, and a printable aig-compliance-report.html is still written. Opening a history row shows that run again on the page. The terminal then prints the name, the verdict, the report path, and a single completion line.
Model calls go directly to the endpoint you enter, with a concurrency setting on the page. An infrastructure address is scanned through aig-infra-scan's scan.
License
Apache License 2.0. Redistribution must state "Based on Tencent Zhuque Lab AI-Infra-Guard" and link to https://github.com/Tencent/AI-Infra-Guard. See NOTICE.
Metadata
Release files for aig-compliance 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| aig_compliance-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for aig_compliance-0.1.0-py3-none-any.whl |
Release files / aig_compliance-0.1.0-py3-none-any.whl
| Download URL | aig_compliance-0.1.0-py3-none-any.whl |
|---|---|
| Size | 60.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
4eb3a5b11d9bb0c9cd2877cab06988cebfc7a72785f12c20f2a689ffaee43dd5
|
|
BLAKE2b-256 checksum How to use checksums |
b0b0b19c0d5d0fae83804edab88725c66fb12fd4dd49e64e087f0554789e7685
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Security for aig-compliance 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.