Vertex AI Search lets developers tap into the power of Google's foundationmodels, search expertise, and conversational AI technologies to createenterprise-grade generative AI applications.As an early-stage technology, itsevolving capabilities and uses create potential for misapplication, misuse, andunintended or unforeseen consequences. For example, Vertex AI Search cangenerate output that you don't expect, including text that's offensive,insensitive, or factually incorrect.

Given these risks and complexities, Vertex AI Search is designed withGoogle'sAI Principles in mind. However, it isimportant for developers to understand and test their models to deploy safelyand responsibly. To aid developers, Vertex AI Search has built-in safetyfilters to help customers block potentially harmful outputs within their usecase. For more information, seeSafety settings for Vertex AI Search.

When Vertex AI Search is integrated into a customer's unique use case andcontext, additional responsible AI considerations and model limitations might needto be considered. We encourage customers to exercise fairness, interpretability,privacy, and securityrecommendedpractices.

Safety settings for Vertex AI Search

Vertex AI Search offers SafeSearch filters to filter explicit websitesearch results. Explicit results might include content like pornography, violence,and gore. SafeSearch filtering is not available for private content.

Summarization (both for website search and private content) filters outresponses such as derogatory, sexually explicit, toxic, or violent information.Additionally, the summarization response contains safety attributes, whichinclude "harmful categories" and topics that might be considered sensitive.

Website search

SafeSearch can be applied to public content indexed byVertex AI Search, such aswebsites. When SafeSearch is on,Vertex AI Search helps filter out explicit content in a customer'sVertex AI Search results for all queries across images, videos, andwebsites. While those filters are not 100% accurate, we continuously evaluateand update filtering capabilities.

In certain contexts and use cases, customers might require access to a wide rangeof information, even if it includes explicit content, to gather comprehensiveinsights and results from Vertex AI Search. Customers can enable anddisable safety filters using the API by settingsafe_searchto "true" or "false".

To enable SafeSearch for Vertex AI Search, use the following APIcall:

curl -X POST\ -H "Authorization: Bearer $(gcloud auth application-default print-access-token)" -H "Content-Type: application/json"\"https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/dataStores/DATA_STORE_ID/servingConfigs/default_config:search"     -d '{      "query": { "input": "FREE_TEXT"},  "serving_config": "SERVING_CONFIG_ID",  "safe_search": "true",}'

To enable SafeSearch for Search with follow-ups in Vertex AI Search,use the following API call:

curl -X POST \-H "Authorization: Bearer $(gcloud auth print-access-token)" \-H "Content-Type: application/json" \"https://discoveryengine.googleapis.com/v1beta/projects/PROJECT_ID/locations/global/collections/default_collection/dataStores/DATA_STORE_ID/conversations/CONVERSATION_ID:converse" \-d '{  "query": { "input": "FREE_TEXT"},  "serving_config": "SERVING_CONFIG_ID",  "safe_search": "true",}'

Thesafe_search setting applies to search results and also toVertex AI Search outputs like summarization (which is based on the searchresults).

Vertex AI Search outputs, such as summarization, are assessed against a list ofsafety attributes which filter potentially harmful content even whensafe_search is turned off.

Summarization for private content and website search

Vertex AI Search outputs, such as summarization, are assessed against a list ofsafety attributes, which filters potentially harmful content such as derogatory,sexually explicit, toxic, or violent information, even whensafe_search forwebsite search is turned off. Additionally, you can filter out responses whichinclude "harmful categories" and topics that might be considered sensitive byusingsafetyAttributes in the Vertex AI Search API response.

Fallback responses

If the summarization model responds to a request with an empty string or ageneric response, it means that either the input or the output is triggering asafety filter. In the Vertex AI Search API response,SummarySkippedReason is set toBLOCKED_DUE_TO_POTENTIAL_POLICY_VIOLATION ifa potential policy violation is detected.

If you think that a safety filter is being inappropriatelytriggered,report this issue using theGoogle Cloud console.

Additional resources

• Learn more about Google's recommendations forResponsible AI practices
• Read our blog,A shared agenda for responsible AI progress