About this site

tampered.app is built and paid for by OneKey, which makes hardware wallets. Two of the three brands covered here are our competitors, so we commit to the following:

The same rules for every brand. Ledger, OneKey and Trezor devices go through the same prompt, the same verdict rules and the same wording. OneKey devices get no softer treatment.

No sales pitch. The site never suggests switching brands and carries no ads. The one commercial link is the OneKey shop link in the OneKey guide: it carries Yishi’s referral code, and it is labelled where it appears.

Official checks first. Every guide sends you to the manufacturer’s own check and support team.

In October 2026 a reseller was found selling a hardware wallet with an implant added to its circuit board. The official checks prove that a device’s security chip is genuine, but they can’t tell whether something has been added next to it. The only way to see that is to open the device and look. This site helps people who aren’t hardware engineers do that.

1. Your browser shrinks each photo to at most 2048 pixels and re-encodes it, which removes metadata such as GPS location.

2. The photos go to a Cloudflare Worker, which sends them to an open-weight vision model on Cloudflare Workers AI, together with the chips a genuine board of your model should carry and the physical signs found in real tampered devices.

3. The model returns structured observations: photo quality, whether the board matches the model, every piece of text it can read, the chips it recognises, and each anomaly with its location.

4. Fixed rules decide the result, not the model. Any serious finding means “Something doesn’t belong”. Any uncertainty means “Take a closer look”. Text that names a cellular radio part counts as a serious finding even if the model missed it. Blurry photos are never judged. The best possible result is “Looks clean”, always with the reminder that photos can’t prove a device is safe.

It can’t see firmware or software, the inside of a chip, or parts under shields, the screen or the battery. It can’t tell a counterfeit chip from a genuine one if both carry the same markings. The model can miss things and can raise false alarms. Treat the result as one of three checks, never as a certificate.

Photos are processed in memory to produce your report. We don’t store them, log them or use them for training. Cloudflare processes them to run the model and states that Workers AI doesn’t use customer content to train models. For each check we log only the device model, the result, the number of findings and the duration. To keep costs under control, each network gets a few checks a day: we count them under a one-way hash of the IP address made with a random value that is replaced every day, and the counts are deleted daily. The text of a report (never the photos) is cached for up to 7 days so the same photos are not checked twice; only someone holding the same photos can get it back. If you share a result, the image you create (your board photo with the result stamp) is published at a tampered.app link for 30 days; nothing is published unless you press the share button. We keep no IP addresses. The site sets no cookies and loads no analytics or third-party scripts. Your language choice is kept in your browser’s local storage.

To report a tampered device, contact the manufacturer’s support team first. For problems with this site, or to share photos of a confirmed tampered device so the reference list can improve, write to OneKey at help.onekey.so.