A student ID that never crosses the Wi-Fi in the clear.

If a classroom attendance page runs over plain http on the school Wi-Fi, every student ID typed into it crosses the air as readable text, and anyone on the same network with a free packet-capture tool can collect the whole class list.

The goal was a system where the ID is unreadable in transit, and where that claim can be checked by anyone with a capture rather than taken on trust.

Two devices and one shared network. The middle lane of the diagram is the Wi-Fi: everything written there is what an attacker can record, and none of it is the student ID.

• The encryption has to happen in the browser. Encrypting on the server would be pointless: the plain ID would already have crossed the network.
• The private key never leaves the board. The key export asks only for n and e and passes nothing for p, q and d, so the secret half is never copied out, turned into text, or sent.
• The RSA math is written out by hand. The browser’s Web Crypto encryption functions only work on secure (https) pages, and the ESP32 serves plain http, so the page computes c = m^e mod n itself with JavaScript BigInt. Its random padding still comes from crypto.getRandomValues, which works on http.
• Random padding is required, not decoration. Student IDs are short and predictable, so without it an attacker could encrypt every possible ID and compare. PKCS#1 v1.5 padding makes every ciphertext different.
• The return trip leaks too. The live table comes back over the same Wi-Fi, so IDs are masked to their last two digits.
• The timestamp comes from the server. The ESP32 stamps each record from its own internet-synced clock, so a client cannot fake when it checked in.

Every claim was checked against a Wireshark capture of real submissions, the same recording an attacker on the network would make. The weakness the tests found is reported alongside the passes.

RSA as used here gives confidentiality, not freshness. The ESP32 accepts any ciphertext that decrypts, however many times it has seen it, which is why the replay test succeeded.

The proposed fix is a one-time nonce. Before each submission the page asks the ESP32 for a random number it will accept once, and encrypts it together with the ID. A replayed request carries a nonce that is already spent, and the attacker cannot wrap a fresh one around the old ciphertext without knowing the ID.

The other gaps are written down rather than hidden: nobody is authenticated, so anyone who knows a classmate’s ID can submit it; over plain http the browser cannot prove it is talking to the real ESP32, the most serious gap; records live in RAM and vanish on a restart; and whoever holds the USB cable sees full IDs on the serial console.