> ## Documentation Index > Fetch the complete documentation index at: https://docs.tuteliq.ai/llms.txt > Use this file to discover all available pages before exploring further. # KOSA Compliance > How Tuteliq covers all nine Kids Online Safety Act harm categories The **Kids Online Safety Act (KOSA)** requires covered platforms to prevent and mitigate specific categories of harm to minors. Tuteliq provides API endpoints that map directly to each of the nine KOSA harm categories, giving you a single integration point for full compliance coverage. ## Coverage Matrix | Harm Category | Tuteliq Endpoint(s) | Status | | --------------------- | ----------------------------------------------- | ------- | | Eating Disorders | `/safety/unsafe` | Covered | | Substance Use | `/safety/unsafe` | Covered | | Suicidal Behaviors | `/safety/unsafe` | Covered | | Depression & Anxiety | `/safety/unsafe` + `/analysis/emotions` | Covered | | Compulsive Usage | `/safety/unsafe` | Covered | | Harassment & Bullying | `/safety/bullying` + `/safety/coercive-control` | Covered | | Sexual Exploitation | `/safety/grooming` + `/safety/unsafe` | Covered | | Voice & Audio Threats | `/safety/voice` | Covered | | Visual Content Risks | `/safety/image` | Covered | ### Extended Coverage (Beyond KOSA) Tuteliq also provides detection for harm categories beyond the core nine KOSA requirements: | Category | Tuteliq Endpoint(s) | Notes | | -------------------------- | ------------------------------------ | ---------------------------------------------------------------- | | Social Engineering | `/fraud/social-engineering` | Pretexting, impersonation, urgency manipulation | | App & Download Fraud | `/fraud/app-detection` | Fake apps, malicious links, clone distribution | | Romance Scams | `/fraud/romance-scam` | Love-bombing, financial exploitation | | Money Mule Recruitment | `/fraud/mule-recruitment` | Laundering recruitment targeting minors | | Gambling Harm | `/safety/gambling-harm` | Underage gambling, addiction patterns | | Coercive Control | `/safety/coercive-control` | Isolation, financial control, surveillance | | Vulnerability Exploitation | `/safety/vulnerability-exploitation` | Cross-endpoint vulnerability profiling | | Radicalisation | `/safety/radicalisation` | Extremist rhetoric, recruitment patterns | | Age Verification | `/verification/age` | Document-based age assurance (Beta — Pro tier) | | Identity Verification | `/verification/identity` | Document + liveness identity confirmation (Beta — Business tier) | ## Category Details ### Eating Disorders The `/safety/unsafe` endpoint detects promotion of disordered eating, pro-anorexia and pro-bulimia content, dangerous diet challenges, and body-shaming language that could trigger eating disorders in minors. ### Substance Use Content promoting or normalizing drug use, alcohol consumption by minors, vaping, and substance abuse challenges is flagged by `/safety/unsafe` under the `substance` sub-category. ### Suicidal Behaviors The `/safety/unsafe` endpoint identifies suicidal ideation, self-harm instructions, and content that glorifies or normalizes self-harm. Detections in this category are always elevated to `critical` severity to enable immediate intervention. Alerts in the suicidal behaviors category should trigger your most urgent response workflow. Consider routing these to human moderators with mandatory response SLAs. ### Depression & Anxiety Tuteliq uses a two-endpoint approach for this category. The `/safety/unsafe` endpoint catches content that promotes hopelessness or emotional distress. The `/analysis/emotions` endpoint provides nuanced emotional tone analysis, detecting sustained patterns of anxiety, despair, or emotional manipulation in conversations. ### Compulsive Usage The `/safety/unsafe` endpoint detects dark patterns, engagement manipulation, and content designed to exploit compulsive usage behaviors in minors, such as artificial urgency, streak mechanics framing, and addictive loop language. ### Harassment & Bullying The dedicated `/safety/bullying` endpoint is purpose-built to detect direct insults, exclusionary language, coordinated harassment, intimidation, and cyberbullying patterns across text conversations. ### Sexual Exploitation This category uses two endpoints for layered detection. The `/safety/grooming` endpoint identifies predatory behavior patterns such as trust-building, isolation tactics, boundary testing, and secret-keeping. It also returns a per-message risk breakdown (`message_analysis`) showing how tactics escalate across the conversation — useful for incident timelines and audit trails. The `/safety/unsafe` endpoint catches explicit sexual content, sextortion language, and exploitation material references. ### Voice & Audio Threats The `/safety/voice` endpoint accepts audio files and returns transcription with safety analysis. It detects verbal harassment, threats, grooming in voice messages, and other harmful audio content. For real-time monitoring, see [Voice Streaming](/voice-streaming). ### Visual Content Risks The `/safety/image` endpoint analyzes images for inappropriate visual content including explicit material, violent imagery, self-harm depictions, and other content that poses risks to minors. ## Age-Calibrated Severity All Tuteliq safety endpoints accept an optional `age_group` parameter within the `context` object. When provided, severity scores are calibrated to the developmental stage of the minor. Content that may be flagged as `medium` for a 16-year-old could be elevated to `high` or `critical` for an 8-year-old. ```json theme={"dark"} { "text": "let's play a secret game, just between us", "context": { "age_group": "under-10" } } ``` Age calibration is applied automatically across all harm categories when the parameter is present. This helps platforms implement age-appropriate safety measures as recommended by KOSA. ## Multilingual KOSA Coverage All KOSA harm categories are detected across 27 languages: English (stable), plus Spanish, Portuguese, Ukrainian, Swedish, Norwegian, Danish, Finnish, German, French, Dutch, Polish, Italian, Turkish, Romanian, Greek, Czech, Hungarian, Bulgarian, Croatian, Slovak, Lithuanian, Latvian, Estonian, Slovenian, Maltese, and Irish (beta). Language is auto-detected — no explicit parameter is required. Each language includes culture-specific analysis guidelines for local slang, idioms, harmful terms, grooming indicators, self-harm coded vocabulary, and filter evasion techniques, ensuring that detection quality is maintained across linguistic contexts. ## Implementation Guidance To achieve full KOSA coverage, integrate the following endpoints into your content pipeline: Route user-generated text through `/safety/unsafe`, `/safety/bullying`, and `/safety/grooming`. This covers eating disorders, substance use, suicidal behaviors, compulsive usage, harassment, and sexual exploitation. For broader coverage, use `/analyse/multi` to fan-out to multiple detection endpoints in a single call. For platforms handling financial interactions or vulnerable users, add `/fraud/social-engineering`, `/fraud/romance-scam`, `/fraud/mule-recruitment`, `/safety/gambling-harm`, `/safety/coercive-control`, and `/safety/radicalisation` to your pipeline. Include `/safety/vulnerability-exploitation` to get a cross-endpoint vulnerability modifier that adjusts severity across all results. For messaging and social features, use `/analysis/emotions` to detect sustained patterns of depression and anxiety in conversations. Send audio messages and voice chat recordings to `/safety/voice`, or use the [WebSocket streaming endpoint](/voice-streaming) for live monitoring. Route uploaded images and profile pictures through `/safety/image` to cover visual content risks. Configure [webhooks](/webhooks) for `safety.critical` and `safety.high` events to ensure immediate response to the most severe incidents.