Product features

After the data such as images, videos, and text submitted by users is reviewed in AI Guardrails, will the data be deleted?

For images, videos, and text that are submitted using AI Guardrails API, AI Guardrails retains the images, videos, text, and the machine-assisted moderation results for seven days. During this period, you can query the data and results. After seven days, the data and results are deleted. For live streams and voice streams that are submitted using AI Guardrails API, temporary addresses are generated for media segment files during the review process. These temporary addresses are valid for 30 minutes. After 30 minutes, the addresses become invalid, and the media segment files are deleted.

If you want to retain the machine-assisted moderation results and media segment files, you must save them before they are deleted. If you have activated Alibaba Cloud Object Storage Service (OSS), we recommend that you enable the feature of storing evidence in OSS buckets.

What are the features of AI Guardrails?

AI Guardrails provides AI Guardrails API, OSS violation detection, and site detection features that apply to various scenarios.

For more information, see Features.

What is a custom text library of AI Guardrails?

AI Guardrails supports custom text libraries of diverse types. You can use custom text libraries to ensure that moderation results meet specific business needs.

For more information, see Manage custom text libraries.

Can I customize configurations for image moderation in AI Guardrails?

Yes. AI Guardrails supports custom image libraries to make detection results more relevant to your business.

For more information, see Custom image libraries.

How to configure human review for AI Guardrails?

The AI Guardrails console displays the moderation results that are returned by AI Guardrails API. You can use the human review feature to review the machine-assisted moderation results as needed.

For more information, see Review machine-assisted moderation results.

Why does not the custom text library that I configured in AI Guardrails take effect?

Custom text libraries support two types: keywords and similar text. They also support exact match and fuzzy match methods. Check the limits for custom text libraries to verify that your configuration meets the requirements. For example, similar text must contain more than 10 characters. If you use a custom text library in a AI Guardrails API scenario, you must apply the library to the corresponding business scenario (Biztype) for it to take effect.

For more information about how to use custom text libraries, see Manage custom text libraries.

Can AI Guardrails detect spelling or grammatical errors?

AI Guardrails does not detect typos or syntax errors.

What politics-related content does AI Guardrails detect?

Terrorist content detection lets you moderate objects such as images for terrorist content, including bloody content, explosion and smoke, special costumes, logos, weapons, political content, violence, crowds, parades, car accidents, flags, and landmarks. You can define the moderation scope for your business by customizing policies for machine-assisted moderation.

For more information, see Customize policies for machine-assisted moderation.

内容安全发现违规，是否可以保留违规证据？

使用内容安全API时，若您希望保存机审识别的结果证据，您可以开启证据转存功能。证据转存与阿里云对象存储OSS配合使用，支持将视频、语音、图片检测中识别到的违规、疑似、正常内容转存到您指定的OSS存储空间，并返回转存文件的OSS URL链接。

For more information, see Detection Results.

本地图片、视频、语音文件如何进行内容安全检测？

强烈建议您使用阿里云对象存储OSS服务，将本地文件存储在云上存储空间，获取公网可以访问的URL后再提交内容检测。

更多信息，请参见什么是对象存储OSS。

What does the rate parameter mean in image moderation?

We recommend that you determine whether an image contains violations based on the values of the suggestion and label parameters instead of the value of the rate parameter. The rate parameter indicates only the confidence level that the AI Guardrails model generates for an image. This parameter does not reflect the risk level of an image.

For more information about the parameters, see Synchronous detection.

Can I export a custom text library from AI Guardrails?

Yes, you can export multiple terms from a custom text library at a time in the AI Guardrails console.

For more information, see Manage custom text libraries.

When does human review in the AI Guardrails console take effect?

You can use the human review feature to review machine-assisted moderation results in the AI Guardrails console. The human review results take effect in real time. The review results of images and text are automatically added to sample libraries. The newly added samples in the sample libraries take about 15 minutes to be effective.

For more information, see Review machine-assisted moderation results.

How long does it take for a custom image or text library to take effect in AI Guardrails?

AI Guardrails supports custom image libraries. You can use custom image libraries to manage the images that you want to block or allow. You can add and remove image samples in custom image libraries. All these operations take effect about 15 minutes after they are performed. AI Guardrails also supports custom text libraries. You can use custom text libraries to manage the text that you want to block or ignore. You can add and remove text patterns and terms in custom text libraries. All these operations take effect about 15 minutes after they are performed.

For more information, see Manage custom text libraries.

内容安全图片审核会对图片进行自动冻结吗？

内容安全不会自动冻结。阿里云内容安全服务只进行检测并将检测结果反馈给用户，不会查看用户数据，更不会对用户数据做任何处理。 如果您希望对检测违规图片进行冻结处理，建议使用OSS违规检测，并开启OSS违规检测的冻结功能。

For more information, see Create an inventory scan task.

Why am I unable to receive callback notifications after a callback URL is specified in AI Guardrails?

The AI Guardrails API and OSS violation detection both support callback notifications. To receive notifications, you can create a message notification in the AI Guardrails console, select the message types, such as machine review results, self-service or manual review results, and then associate the notification with a business scenario (Biztype).

If you still cannot receive a callback notification from AI Guardrails after the preceding operations, we recommend that you check whether your server can respond to the POST requests sent to the specified callback URL. Make sure that no 403 or 502 error occurs when the callback URL is requested. You can also set the callback parameter in an API request to specify a callback URL. After the AI Guardrails API operation is called, AI Guardrails sends a callback notification to the specified callback URL.

For more information, see Configure notifications.

使用OSS增量扫描，没有回调配置的通知地址？

内容检测API和OSS违规检测均支持回调通知，如果您已经正确设置了消息通知，但是调用内容安全后却没有收到回调，建议检查回调地址是否能够响应POST请求，避免请求回调地址时出现403或者502等错误。

For more information about callback notifications for OSS violation detection, see Configure message notifications.

Why am I unable to moderate a long image after I add long image samples to a custom image library?

AI Guardrails cannot directly moderate images whose height exceeds 400 pixels or images whose aspect ratio exceeds 2.5. For a long image whose height or aspect ratio exceeds the upper limit, AI Guardrails crops the long image and then moderates it. As a result, the cropped long image does not hit the long image samples added to a custom image library or feedback-based image library. Therefore, the long image cannot be moderated.

For more information about the interval and maxframes parameters for long image moderation, see Synchronous detection.

Why does not a custom policy for machine-assisted moderation configured in the AI Guardrails console take effect when I call the AI Guardrails API?

After you customize a policy for machine-assisted moderation or add a sample to a custom sample library in the AI Guardrails console, the custom policy or sample takes 15 minutes to be effective. We recommend that you try again later. In addition, after you customize a policy for machine-assisted moderation based on a business scenario, you must specify the business scenario in an API request for content moderation. Then, the corresponding moderation policy takes effect.

For more information, see Customize policies for machine-assisted moderation.

How to connect to the AI Guardrails internal network address?

If the Elastic Compute Service (ECS) instance on which your service is deployed resides in the same region as AI Guardrails, you can access AI Guardrails using an internal endpoint.

For example, if the ECS instance resides in the China (Beijing) region in which AI Guardrails is deployed, you can use the internal endpoint green-vpc.cn-beijing.aliyuncs.com to access AI Guardrails. Compared with a public endpoint, an internal endpoint provides faster access to AI Guardrails. However, if the ECS instance does not reside in the same region as AI Guardrails, you can use only a public endpoint to access AI Guardrails.

For more information about the endpoints of AI Guardrails, see Endpoints.