This AI & Media Processing Notice ("Notice") explains how TruSight uses artificial intelligence, media processing, OCR, search, transcription, document analysis, identity-registry lookups, legal-record screening, writing detection, and related technologies to generate informational outputs.
This Notice forms part of our Terms of Use and Privacy Policy. By using TruSight, you acknowledge this Notice.
1. Purpose
TruSight helps users evaluate claims and media they choose to submit. The Service combines:
- Artificial intelligence and machine learning models
- OCR and computer vision
- Web search and evidence retrieval
- Transcription and audio analysis
- URL and security checks
- Metadata and media processing
- Document classification and authenticity estimation
- Identity and document registry lookups (where configured)
- Legal-record screening of reported FIR and court proceedings (where configured; identity Document Check only)
- Writing-originality / AI-text detection (where configured)
Outputs are intended to assist human review — not replace human judgment.
2. What the Service processes
Depending on features you use, TruSight may process:
- Text claims you type or paste
- Screenshots and images
- Video files you upload
- Supported video links (e.g. YouTube, Shorts, reels)
- OCR text extracted from images, frames, or document pages
- Photographs, scans, PDFs, and Word files submitted for Document Check
- Audio and transcripts from videos or user-initiated screen recordings
- Content sent via Share Extension or similar share flows (URL, image, or video)
- URLs found in submitted content
The Service may generate Trust Rings, authenticity or confidence scores, summaries, source references, scam-risk signals, fact-check style reports, document authenticity bands, identity-check summaries, writing-originality estimates, and other AI outputs.
You control what you submit. Do not submit content you do not have the right to analyze.
3. Artificial intelligence systems
TruSight uses AI technologies from internal systems and third-party providers. These may include large language models, vision models, OCR engines, search APIs, transcription services, and media analysis tools.
Providers used may change over time. We do not guarantee the continued availability of any specific provider or model.
Typical provider categories include:
- AI models — e.g. OpenAI, Google Gemini, Anthropic, xAI (as configured)
- Search — e.g. Tavily, Brave Search, Perplexity (as configured)
- Media / video — e.g. Supadata, Hive, YouTube API (as configured)
- Document OCR / document AI — e.g. Google Document AI or similar (as configured)
- Identity / document registries — e.g. Surepass-class verification APIs (as configured)
- Legal-record / FIR and court screening — e.g. Deepvue-class APIs (as configured for identity Document Check; name and address only, not ID numbers or images)
- AI writing detection — e.g. Winston AI or similar (as configured)
- URL security — e.g. Google Web Risk for phishing/malware URL checks
- Hosting — cloud infrastructure that processes API requests
- Push delivery — e.g. Apple Push Notification service on iOS
Submitted content may be sent to these providers only as needed to generate results.
4. Trust Rings
Trust Rings are visual indicators generated by automated systems. They:
- Are not statements of fact
- Are not proof of truth or falsity
- Are not legal, journalistic, or investigative findings
- Are not guarantees
A high Trust Ring does not mean information is true. A low Trust Ring does not mean information is false.
Trust Rings may consider signals such as source availability, claim matching, OCR quality, search results, context, and AI analysis. Methodology is proprietary and may change without notice.
5. Authenticity scores and media signals
Authenticity or footage scores are algorithmic estimates — not forensic analyses, expert opinions, or legal determinations.
They may consider visual indicators, metadata, transcripts, frames, and model outputs. A score does not prove that media is genuine or manipulated.
The Service may not detect all deepfakes, synthetic media, edits, or deceptive content. Failure to flag manipulation does not prove authenticity; a flag does not prove fraud.
5A. Document authenticity and identity checks
Document Check may:
- 1. OCR text from photos, scans, or PDF pages (including scanned PDFs that have little native text)
- 2. Classify the capture (identity document, issued document such as an invoice or certificate, writing sample, claim-bearing content, or not a document)
- 3. Run local format and checksum checks on identifiers where rules exist
- 4. Call third-party registry / verification APIs with extracted identifiers and/or prepared images (for example passport, PAN, Aadhaar, driving licence, voter ID, vehicle RC, GSTIN) where configured
- 5. For identity documents, where configured, call a legal-record screening API with name and address (and father’s name or date of birth only if already extracted) to look up reported FIR and court proceedings. This is not the same as issuer-registry verification and does not send ID numbers or images
- 6. Use AI models and visual signals to estimate consistency, tampering risk, or issuer context
- 7. For writing-focused paths, use stylometry / classifiers and optional third-party AI-text detectors, plus web reuse signals
What outputs mean
- Trust Rings / bands (for example Authentic, Likely authentic, Uncertain, Fake) are estimates, not certificates.
- A registry match does not equal DigiLocker, NAD, university, employer, or government clearance.
- Expired or inactive registry status may still indicate a real record; it does not automatically mean “fake.”
- Missing registry coverage is incompleteness, not proof of fraud.
- Legal-record hits are reported proceedings, not proof of guilt. An empty result is not proof that no FIRs or cases exist in India. Coverage is not every FIR or court record in India.
- Writing-originality signals are not proof of authorship, plagiarism findings, or academic misconduct determinations.
Always verify important documents with the issuer or an official channel.
6. Scam and URL risk signals
For text messages, links, and similar content, the Service may apply automated heuristics and URL security lookups (such as Google Web Risk) to help identify possible phishing or malicious links.
These signals are informational only. They are not guarantees that a message is safe or fraudulent. Always verify through official channels before clicking links or sharing credentials.
7. AI hallucinations and errors
AI systems may produce inaccurate, fabricated, incomplete, outdated, biased, or misleading outputs ("hallucinations"). This may affect summaries, citations, source references, explanations, and scores.
Do not assume AI outputs are accurate. Independently verify before acting, sharing, or publishing.
8. OCR processing
OCR may extract text from screenshots, images, documents, or video frames. OCR can misread characters, numbers, names, dates, URLs, or symbols — especially with poor image quality, cropping, or stylized fonts. Verify extracted text independently.
9. Screenshot and image analysis
Screenshot and image analysis may use OCR, visual analysis, and AI interpretation. The Service may lack full context (conversation thread, surrounding page, satire, edits). Results may be wrong due to missing context or image quality limits.
10. Sherlock screen capture and Share
Optional Sherlock features may capture screenshots or short screen recordings after your explicit action and system permission approval.
Android may require display-over-other-apps (floating assist), media projection, and microphone if you enable audio.
iOS may use Broadcast Upload / Control Center recording, Action Button or similar screenshot flows, photo/video library access, microphone if you enable audio, and the Share Extension to send content from other apps.
We do not continuously monitor or record your screen. You are responsible for ensuring you may lawfully capture and analyze content, including privacy, platform terms, and third-party rights.
11. Video and link analysis
Video analysis may use transcripts, frame sampling, OCR, metadata, and search-based verification. Analysis may be limited by quality, editing, compression, missing audio, blocked captions, or platform restrictions.
For video links, the Service may fetch available metadata, transcripts, or previews. Not all information about a linked video may be accessible.
We may not analyze every frame of a video.
12. Fact-checking and reputation limitations
TruSight is not a fact-certification authority, publisher, or reputation-rating agency. Claim assessments may be incomplete, outdated, or based on limited sources.
Outputs relating to people, organizations, governments, or public figures are automated — not factual findings about credibility, honesty, or intent. We are not liable for reputational harm or publication decisions based on outputs.
Political or election-related content may be processed for informational analysis only. We do not endorse any candidate, party, or ideology.
13. Source references
The Service may cite or link to external sources. Sources may be inaccurate, unavailable, outdated, or conflicting. Source ranking or prominence is automated and is not an endorsement or reliability guarantee. Evaluate sources yourself.
14. Automated processing and high-risk uses
Outputs are generated by automated systems to assist users. They are not intended to produce legal, medical, financial, employment, immigration, KYC, or similarly significant effects without human review.
The Service is not designed for healthcare diagnosis, emergency response, law enforcement, judicial decisions, critical infrastructure, child-protection investigations, regulated identity verification, or other safety-critical uses.
Organizations (businesses, media, schools, governments) use the Service as an informational aid only — not as a substitute for professional investigation, legal review, compliance, journalism standards, KYC, or due diligence.
15. Data handling summary
- Submitted content is processed to generate results. We do not intend to keep full raw submissions on our servers after processing, except as described in the Privacy Policy for the anonymized result cache, share links, short-lived analysis jobs, security logs, and legal retention.
- Result cache (Truth Vault): Anonymized analysis results may be stored for up to 30 days and reused for closely matching content (approximately 90% text / 80% image similarity or higher, depending on media type). A cache hit may return a prior result without re-running the full AI pipeline; fresh checks use the same methodology. Sensitive tokens such as OTP codes and phone numbers are redacted where feasible in stored previews. Truth Vault is enabled for all users while active on our servers; there is no in-app opt-out.
- If you create a share link, a limited result summary may be stored so recipients can open it.
- Share / background jobs may retain job metadata and results for a short period (typically up to about 24 hours) so notifications and the app can show completed checks.
- Local history on your device is under your control and can be deleted in the app. Optional encrypted backups stay in your iCloud or Google Drive account.
- The current apps do not use dedicated third-party advertising, analytics, or crash-reporting SDKs for this processing.
16. No professional advice
Nothing generated by the Service is legal, medical, financial, investment, tax, security, or other professional advice. Consult qualified professionals for important decisions.
17. User responsibilities
You are solely responsible for evaluating outputs, verifying information, complying with law, respecting third-party rights, and deciding whether to share or act on results.
18. Changes
We may update this Notice at https://smitten.co.in/media-processing-notice. Continued use after updates constitutes acceptance.
19. Contact
Smitten Technologies Delhi, India Email: info@smitten.co.in Website: https://smitten.co.in
*End of AI & Media Processing Notice*