Privacy
How it works & privacy
ScamScan is a trust tool for checking a recruiter approach before you reply. This page explains what we do with what you paste — including that the message text is processed by Anthropic’s Claude API.
Short version
- Your message text — and any screenshot you attach — is sent to Anthropic (Claude) once per scan so we can classify scam wording patterns.
- ScamScan does not store the message, the screenshot, or a history, and there are no accounts.
- We never scrape LinkedIn — profile answers come from what you see on your own screen. Our server does run automated safety checks on the application link (redirects and blocklists) without ever loading the page content, and nothing is opened on your device.
What happens when you scan
- Your pasted message (and screenshot, if you attach one) is sent once to Anthropic’s Claude API, which classifies wording patterns only (payment asks, crypto, platform-switching, etc.). Claude is not asked to invent facts about companies or people.
- Deterministic checks run in our code: domain match / lookalike detection, public domain age via RDAP, link destination checks (shorteners, redirect chains, Google Safe Browsing) done from our server without loading page content, and — if you answer them — LinkedIn profile questions.
- The verdict is computed in code from those signals. A result below high risk never gets more than medium confidence, and we never call an approach “definitely safe.”
What we never do
- We don’t create accounts, save history, or keep a database of messages.
- We don’t scrape LinkedIn or open links you paste.
- We don’t assert that a person “is a scammer” — we assess the interaction you showed us.
- If you leave feedback (yes/no + a short note), we log a count and a structured log line only — never with the original message.
What is retained briefly
Rate-limit counters (per IP and global) live for about a minute so the Anthropic budget can’t be drained. When Upstash Redis is configured in production, those counters are shared across servers and expire automatically. Aggregate scan counts and verdict breakdowns may appear in operator logs and a token-gated stats endpoint — still without message content.
We count visits ourselves instead of using a third-party analytics service: daily page totals, an anonymous estimate of unique visitors (a one-way hash that changes every day), and which sites referred traffic. No cookies, no advertising identifiers, and no record of any individual’s browsing.
Anthropic processes the message under their API terms for the duration of that single request. We don’t ask them to store it for training of this product.
LinkedIn checks
When the approach arrived on LinkedIn, we ask you to read About this profile and report what you see. That keeps legal and technical risk low — LinkedIn blocks scraping, and we never need credentials or automated access to your account.
Questions about this page? Treat ScamScan as an aid, not legal advice — and verify any role through channels you find yourself.