Don't trust a QR until it's been cleared.

Quishing is QR-code phishing: an attacker prints, stickers, emails, or DMs a QR code that, when scanned, opens a credential-harvesting page, a wallet-drainer transaction, an open WiFi trap, or an Android intent that sideloads malware. The QR itself is just an image, so email-link filters and browser warnings never see it until the victim's phone has already opened the destination. The defense has to happen at scan time, before the phone follows the link.

Three differences. The attack vector is physical or visual, a sticker over a parking-meter QR, a printed flyer impersonating an MFA enrollment, a restaurant menu QR replaced overnight, so it bypasses email gateways entirely. Victims trust QR codes more than they trust links in email; a QR feels like a destination chosen by whoever printed the surface. And dynamic QR codes that route through a shortener can be repointed at a phishing destination after the printed asset has been distributed, so a QR that was safe at print time can become hostile months later. Static link-scanners answer "is this URL malicious right now", not "who can change where this points".

Don't point your phone's native camera at it, that opens the destination immediately. Instead, open check.qr.abundera.ai on your phone, scan the QR through the in-page camera (decoding happens locally; the image never leaves your device), and read the verdict. We walk every redirect hop, classify whether the destination is controllable by a third party after the QR was printed, and check reputation against Google Safe Browsing and other aggregators. Cleared verdicts are safe to open; Caution and Do not proceed verdicts tell you why.

Parking-meter and EV-charger stickers that overlay the legitimate QR with one pointing to a credit-card-harvesting page, the most-reported pattern of 2024-2025, observed in Austin, San Antonio, and across the UK. Restaurant menu QRs swapped to phishing pages overnight by an attacker physically replacing the table tent. Corporate MFA-enrollment flyers in office bathrooms that look official but enroll the attacker's device. Wedding-invitation QRs distributed months before the event, where a shortener-account compromise lets an attacker repoint thousands of printed cards. Crypto-payment QRs at point-of-sale terminals overlaid with the attacker's wallet address. The common thread: the printed QR looks identical to the safe one.

Existing tools classify whether a URL is currently malicious. We additionally classify whether the destination is controllable by a third party after the QR was printed, a property we call mutability. A clean dynamic QR routed through a shortener is still high-risk for a parking-meter sticker or wedding invitation: the shortener account holder can change the destination at any time. We surface this control-posture as a first-class verdict field alongside the threat-content verdict.

No. The decoded payload travels to our server over HTTPS so we can walk the chain and query reputation databases, that's a functional necessity, not a choice, but it is never persisted. Verdicts are cached by a SHA-256 hash of a per-payload-type discriminator concatenated with a server-held secret salt. The original payload cannot be reconstructed from any cache entry.

qr.abundera.ai is a generator that promises everything-client-side: nothing leaves your device. This safety checker transmits the decoded payload to the server by necessity. We separate the two surfaces so the client-only promise stays clean on the generator domain, and the inverted-privacy-model surface stays clearly labeled here.

Pro tier. Submit a QR for tracking, and we re-walk the chain on a periodic cadence. Email when the redirect targets, terminal destination, or the set of indirection-service operators changes. This catches the most common quishing-in-the-wild pattern: print a clean QR, switch the destination to phishing months later, harvest scans from the printed asset.

Yes, on the Pro tier. The API is RESTful, returns a structured JSON verdict with payload-type, threat-class, mutability, redirect chain, per-hop control attribution, and sub-payload findings. Designed for embedding in wallet apps, mobile security suites, enterprise URL filtering, and corporate Slack / Teams link-preview enrichers.

Not at this time. The classifier is closed-source while the underlying patent work is in prosecution. We may publish reference implementations of disclosed algorithms after grant. The API contract is public and stable.