What runs under the hood.

Detection runs against public-source observations - indexed content from known leak clusters, Telegram bridges, and monitored domains - so creators can investigate what is already circulating under their own handle.

• Indexed sources A discovery pipeline harvests known leak-adjacent sources - including Telegram bridges - on an ongoing basis, computing a perceptual fingerprint for surfaced media and indexing it by handle, domain, cluster, and recency. Monitoring of your enrolled, ownership-verified handles runs against this index.
• OCR on surfaced images When reachable images or frames contain visible text, Sentixx runs OCR as part of the scan. If your creator handle appeared on the image itself - for example as a watermark, username tag, or screen label - the chance of surfacing that copy is materially higher.
• Perceptual fingerprint matching Where visual media is reachable, a perceptual hash is computed and compared against your own enrolled reference fingerprint. Hash distance determines match confidence - not a binary yes/no.
• Repeat cluster tracking Sources are tracked as propagation clusters, not just individual URLs. When the same content appears across multiple nodes, the spread pattern is modelled - useful for understanding velocity and origin proximity.
• What the index does not cover Gated sources, dead links, private channels, and actively-removed pages are outside the observable set. Coverage is honest and stated, not claimed as exhaustive.

Detection and correlation are algorithmically assisted. Decisions are not. Sentixx does not automatically accuse, revoke access, or file reports on your behalf - it surfaces evidence-quality records so you can respond with facts instead of panic.

• What AI does here Pattern matching across indexed sources, perceptual hash comparison, propagation graph generation, confidence scoring. The goal is to reduce noise and surface the candidates worth reviewing - not to replace the reviewer.
• Confidence scoring Every match comes with a numeric confidence estimate and a plain-language statement of what could degrade it: re-encoding depth, crop extent, audio-only correlation, and so on. A score of 0.94 is not a verdict - it is a starting point for review.
• Human review layer The evidence record is built for a person - creator, legal contact, platform trust-and-safety team - to read, assess, and decide what to do next. Sentixx prepares the record; it does not make the call.
• Why this matters Automated accusations cause real harm when they are wrong. False positives on intimate content are especially costly. The design choice to keep a human in the loop is intentional, not a gap in the product.

Sentixx's detection pipeline is designed to recognize content that carries a per-buyer forensic watermark - the same invisible, signal-level mark that sixx.me's protected delivery embeds in every file it sends to a buyer. Watermark embedding, delivery gates, and per-buyer marking are a sixx.me feature, not something Sentixx does on this site. If you want to sell protected, per-buyer-marked drops, that workflow lives on sixx.me; if you want to know what is already circulating under your handle, that is what Sentixx does here.