Documentation

VeriTrust platform guide.

VeriTrust helps teams review suspicious images, messages, emails, and links with a clear, business-friendly result. The platform is designed for fast triage, safer decisions, and consistent review workflows.

Use VeriTrust as a decision-support layer. A result can help prioritize review, but high-impact decisions should still include human judgment and internal policy checks.

Common Use Cases

• Review user-submitted profile photos, campaign images, and public-facing media for possible synthetic or edited content.
• Check suspicious emails, SMS messages, links, or internal reports before employees engage with them.
• Support moderation, compliance, security operations, and trust-and-safety workflows with a repeatable first-pass review.
• Create a simple review path for non-technical teams that need a clear result without reading raw model output.

Image Checks

Use Image Check when you need to review whether a face image may be edited, generated, or otherwise synthetic. Clear, front-facing images usually produce the most useful result.

• Upload a supported image format such as JPG, PNG, or WEBP.
• Use Crop Face when the image includes extra background, multiple people, or a small face area.
• Compare the result with the original context before taking action.
• For sensitive cases, keep a record of the source, review time, and the final human decision.

Message Checks

Use Message Check for emails, SMS messages, URLs, and short suspicious text. Paste the full message when possible so VeriTrust can review context, language patterns, links, and common scam indicators.

• Include sender text, link text, and call-to-action wording when available.
• Avoid submitting passwords, one-time codes, payment details, or private customer records.
• Treat urgent payment requests, credential prompts, and unusual login links as higher priority for manual review.
• Use the result to decide whether to block, escalate, educate the user, or request more information.

Link Intelligence

Use Link Intelligence to review suspicious URLs, shortened links, and domain impersonation patterns before opening them. The MVP analyzes URL strings and surrounding context; it does not fetch webpages or follow redirects.

• Submit only http:// or https:// URLs.
• If a message contains multiple URLs, VeriTrust analyzes the first URL and returns the extracted URL list.
• Indicators include shorteners, IP-address URLs, punycode patterns, suspicious TLDs, missing HTTPS, login/KYC/payment terms, credential terms, brand impersonation patterns, redirect parameters, embedded URLs, and random-looking tokens.
• Use official websites, verified apps, or direct support channels to confirm suspicious links.

Privacy

VeriTrust is designed to minimize unnecessary exposure of sensitive information. Users should only submit content that is needed for review and should avoid pasting secrets or private account data.

• Submit only the image or text required for the check.
• Remove passwords, access codes, customer identifiers, and financial details before review.
• Keep review access limited to authorized team members.
• Follow your organization's retention, consent, and incident-response policies.

Security

VeriTrust should be used as part of a broader security process that includes access control, user education, escalation paths, and audit-friendly workflows.

• Use strong account credentials and avoid shared logins.
• Escalate high-risk results to your security or trust-and-safety owner.
• Do not rely on a single score for account bans, employee discipline, or legal decisions.
• Review suspicious content in a safe environment and avoid opening unknown links directly.

Account Flow

VeriTrust accounts help teams keep review activity organized across workspaces, saved scans, usage, and access controls.

• Use the Auth page to sign in or create a workspace.
• Use the Dashboard page to review account, usage, organization, and recent scan activity.
• Detection pages attach completed reviews to the active workspace when a user is signed in.
• Workspace access should be limited to authorized reviewers and administrators.

Understanding Results

VeriTrust presents a simple verdict, a confidence value, and a risk level. These signals are intended to make review faster and more consistent.

• Verdict: The most likely classification for the submitted content.
• Confidence: How strongly the model score supports the verdict. Strong confidence is not proof.
• Confidence band: Weak, Moderate, or Strong, based on the reported confidence score.
• Risk: Low, Medium, High, or Critical, used for triage and escalation priority.
• Indicators: Rule-based or model-reported signals that explain why content may need attention.

Phishing checks combine model score with deterministic indicators such as urgency, credential requests, OTP or password language, payment wording, account blocking pressure, short links, suspicious domains, attachments, sender identity, and contact methods.

Link Intelligence checks combine URL classifier output with deterministic URL indicators. It does not claim domain age, live reputation, blocklist status, or confirmed maliciousness.

Deepfake checks are based on model score output. VeriTrust does not claim visual forensic findings such as lighting inconsistency or texture artifacts unless a model explicitly provides that evidence.

Model fallback may occur when the selected provider is unavailable. When fallback is used, the report identifies that a backup model handled the scan.

False positives and false negatives are possible. When a result is unclear, collect more context, compare against trusted sources, and route the case to a human reviewer.

View model performance notes for current evaluation status and planned benchmark metrics.

Reports and Exports

Successful scans include report-ready JSON with scan ID, scan type, created date, model information, result summary, evidence or indicators, extracted phishing entities or link URL details when available, and a disclaimer.

• Download JSON reports for audit trails or internal case notes.
• Use Print / Save as PDF for lightweight PDF exports from the browser.
• Copy the result summary when you need a short escalation note.
• Saved dashboard scans can be viewed, downloaded, and printed from scan history.

Reports are AI-assisted assessments. They should support review workflows, not replace manual verification for high-impact decisions.