A sextortion detection API purpose-built for child-safety. Instead of scoring polarity like a sentiment analysis API or single-utterance labels like generic emotion AI, this endpoint performs pre-vulnerability detection across five phases (contact, rapport, leverage, coercion, crisis) with age calibration and evidence spans, so reviewers see why, not just a score.
How this differs from sentiment analysis and generic emotion AI
Sextortion rarely surfaces as a single "negative" message. It is a coercion trajectory across turns where the perpetrator's leverage and the victim's compliance pressure only make sense in sequence.
Dimension
Sentiment analysis API
Generic emotion AI
Tuteliq sextortion detection
What it scores
Polarity: positive / neutral / negative
Emotion labels: joy, sadness, anger, fear…
Five phases + severity + per-tactic categories
Unit of analysis
Single message
Single utterance
Conversation sequence with role/turn structure
Age calibration
None
Multipliers by age group (under-13, 13-15, 16-17, 18+)
Evidence returned
Score only
Evidence spans, phase, referral hints, recommended_action
Best for
Reviews, brand monitoring
UX research, product analytics
Child-safety moderation and law-enforcement escalation
When to use this endpoint
- Use sextortion when coercion, image leverage, or financial demands appear.
- Use grooming for the relationship-building phase that often precedes it.
- Use emotional distress for pre-vulnerability signals in the victim's messages.
- Use synthetic content to check attached media for AI-generated CSAM indicators.
Endpoint
Credits: 5
The five phases we score
Initial approach, often via a compromised or fake account. May include flattery or shared-interest hooks.
Fast-moving intimacy, requests for images or intimate conversation, or claims of prior exchange.
Claim of possession, 'I have the photos', 'I know who your friends are', screenshots of contact lists.
Explicit threats: public exposure, contacting family/school, financial demands. Deadlines and pressure tactics.
Victim shows compliance pressure, panic, self-blame, or self-harm ideation. Escalation-critical.
Request Body
Response
Accuracy & methodology
The sextortion model is trained on sequences with role/turn structure and phase labels, including adversarial and role-play negatives. Evaluation is led by Dr Nicola Harding (clinical criminologist) with clinical reviewers. We deliberately do not publish a single accuracy number: performance is threshold- and age-dependent, and a fair figure for adult platforms differs from one calibrated for under-13 chat. We recommend human-in-the-loop review above medium severity and treat the endpoint as an escalation trigger, not an auto-ban.