Integrate Azure AI Content Safety for Java Apps

This skill covers the Azure AI Content Safety SDK for Java (com.azure:azure-ai-contentsafety:1.1.0-beta.1), used to build content moderation applications. It has two clients: ContentSafetyClient for analyzing text and images, and BlocklistClient for managing custom blocklists - both buildable with either an API-key KeyCredential or DefaultAzureCredential. Analysis scores content across four harm categories - Hate, Sexual, Violence, and Self-harm - on a severity scale (text defaults to 0/2/4/6 of a 0-7 range, image uses a trimmed 0/2/4/6 scale), and analyzeText can be scoped to specific categories or an eight-severity-level output type. Blocklist-aware analysis passes setBlocklistNames and setHaltOnBlocklistHit, and returns any TextBlocklistMatch hits with the blocklist name, item ID, and matched text.

Use it when moderating user-generated text or images in a Java application - detecting hate/sexual/violence/self-harm content, screening against custom blocklists, or building a moderation pipeline with severity thresholds. It's a synchronous, request/response moderation API, not a real-time streaming filter, and blocklist changes take about 5 minutes to take effect, so it isn't suited to moderation rules that must apply instantly.

Configuration is two environment variables, CONTENT_SAFETY_ENDPOINT and CONTENT_SAFETY_KEY. Text analysis takes a string and returns categorized severity scores:

AnalyzeTextResult result = contentSafetyClient.analyzeText(
    new AnalyzeTextOptions("This is text to analyze"));

for (TextCategoriesAnalysis category : result.getCategoriesAnalysis()) {
    System.out.printf("Category: %s, Severity: %d%n",
        category.getCategory(),
        category.getSeverity());
}

Text analysis can also be scoped with setCategories (for example HATE and VIOLENCE only) and setOutputType(AnalyzeTextOutputType.EIGHT_SEVERITY_LEVELS) for finer-grained scores, or combined with setBlocklistNames and setHaltOnBlocklistHit(true) so a match against a named blocklist returns a TextBlocklistMatch with the blocklist name, item ID, and matched text. Image analysis takes either raw bytes (via BinaryData) or a blob URL and returns the same category/severity shape through analyzeImage. Blocklist management returns 201 on create or 200 on update from createOrUpdateTextBlocklistWithResponse, and separate calls add items, list blocklists, get a single blocklist, list or remove items, and delete a blocklist entirely. Errors surface as HttpResponseException with codes such as InvalidRequestBody, ResourceNotFound, or TooManyRequests.

Authenticates via either com.azure.core.credential.KeyCredential or Azure Identity's DefaultAzureCredentialBuilder, and is distributed as the Maven Central com.azure:azure-ai-contentsafety artifact.

Java developers adding content moderation to an application who need harm-category scoring, custom blocklist enforcement, or both - and who can work within the SDK's guidance of requesting only the categories they need (to reduce latency), blocking at severity 4 or above for strict moderation, batching items in parallel for throughput, and caching blocklist results where appropriate.