Back to Writeups
Protocol Analysis & Cryptanalysis

Protocol Analysis of the ASSA ABLOY VERSO CLIQ Key-Cylinder Interface

October 2026 Black-Box Protocol Analysis & Statistical Testing 7 min read
Signal Decoding Packet Format Plaintext Leakage Statistical Testing Relay Timing 2014 vs 2024

Disclaimer & Responsible Disclosure

Published solely for academic study and security research. It describes one CLIQ installation as captured in 2024 (system ID abbreviated as V1004XXX); other installations or firmware may behave differently. No working exploits, key-recovery tools or unlock-emulation code are provided, and no key-recovery or relay attack is demonstrated. This is independent research, not affiliated with or endorsed by ASSA ABLOY or IKON. The author assumes no responsibility for misuse.

Overview

A black-box analysis of the contact-based communication between ASSA ABLOY VERSO® CLIQ keys and lock cylinders, based on 22 unlock sessions from two keys and two cylinders captured with a Saleae logic analyzer between May and August 2024. It covers the signal decoding and complete packet format including the checksum, the five-step message flow and how it has changed since 2014, the fields sent in plaintext, a statistical characterisation of the cryptographic fields, and timing measurements that bound the latency available to a relay attack.

This is part two of a master's project at the University of Rostock. Part one, Forwarding Attack on ASSA ABLOY CLIQ Lock, covers the capture setup and first decoding.

Signal Decoding

The key is the bus master and powers the battery-less cylinder through the contact. A bit '1' is a ~4.3 µs low pulse and a bit '0' a ~13.7 µs low pulse in an ~18.75 µs slot, sent LSB first at about 2.98 V.

While the cylinder computes, the key reads a single status bit from it about every 1.15 ms. An earlier version of the decoder concatenated all bits of a capture, so these busy-poll bits shifted every following byte by 4 bits, and no nonce, encrypted block or MAC was recoverable. All results derived from that decoding have been withdrawn. The new decoder splits the signal at every idle gap longer than 150 µs, re-establishes byte alignment after each one, and keeps busy polls separate.

Result: 308 of 308 packets in the 22 sessions (and 8 of 8 in the fragments) have a consistent length field and a valid checksum. All 22 sessions end in a successful unlock.

Packet Format & Message Flow

Command  (key -> cylinder):  82 00 SEQ CODE LEN DATA[LEN] CHK
Response (cylinder -> key):  84 00 SEQ STS  LEN DATA[LEN] CHK

CODE is 01 identity, 08 nonce request, 0A encrypted block or 80 MAC. STS is 10 busy, 11 OK, or an error (21, 23, 34 observed). CHK is an 8-bit additive checksum: all bytes of a packet, including CHK, sum to 0x00 mod 256. This holds for every 2024 packet and all 84 packets of the 2014 decodes. It is error detection only, which is why earlier searches over CRC-8 parameter spaces found nothing.

Every 2024 session consists of the same 14 packets in five steps:

  • Step 1: Identity, 29-byte format (LEN 0x1D)
    The cylinder refuses it with error 0x34.
  • Step 2: Identity, 27-byte format (LEN 0x1B)
    The key immediately falls back to the older layout. The cylinder replies busy, then sends its own 24-byte identity block.
  • Step 3: Nonce request (CMD 0x08)
    The cylinder generates a fresh 8-byte nonce, different in every session.
  • Step 4: Encrypted block (CMD 0x0A)
    24 bytes that change every session plus 8 zero bytes. The cylinder answers with 24 zero bytes, or with error 0x21 for an unauthorized key. Three wrong-key attempts on 2024-08-23 all passed identity and nonce and failed here, so this is where authorization is decided.
  • Step 5: MAC (CMD 0x80)
    A 21-byte MAC. The cylinder replies 01 01 and releases. A 2014 session rejected at this step (error 0x23) shows the MAC is verified separately.

Change since 2014: in all eight 2014 sessions recorded at the same installation, the key sends only the 27-byte identity and the flow has four steps. The likely explanation is that the keys were updated to a newer protocol version while the cylinders still speak the older one, so the key falls back. Whether a newer cylinder can also be made to fall back is an open question.

What Travels in Plaintext

System ID

Bytes 1–8 of the key's identity packet carry the installation's system ID as ASCII (V1004XXX), echoed back by the cylinder.

Key Configuration

Bytes 16–21 are constant per key (key A 43 0E 00 27 06 04, key B 43 0E 00 3F 0F 04). Bytes 19–20 are the only ones that distinguish the keys, probably a key number.

Key Date & Time

Bytes 22–26 are the key's real-time clock: a day counter from 2000-01-01, minute of the day and second. The dates match every capture date, and the offset to the analyzer's clock is consistent per occasion. Re-programming a key visibly reset its clock by ~51 s.

Cylinder Number

The cylinder's identity reply contains a 16-bit cylinder number (16, 397, and 400 in 2014). Key B was captured at two doors ten minutes apart, and the number changes exactly between them.

Consequence: in every unlock, a device on the contact learns which key was used, at which door, and when to the second — the same information as the system's audit trail. A key's clock offset is also a weak fingerprint of the key.

Cryptographic Fields & Statistical Testing

ASSA ABLOY states that all key-cylinder communication is protected with 128-bit AES. Passive captures cannot confirm this or identify the construction: uniform-looking output says nothing about the algorithm, the 21-byte (168-bit) MAC does not match the output size of any common primitive, and the 24 + 8 byte layout of the encrypted block fits several AES modes.

Over the 22 sessions:

  • Nonce (64 bit): 22/22 unique, one-bit fraction 0.465, mean pairwise Hamming distance 49.4 %, no biased bit positions after Bonferroni correction.
  • Encrypted block (192 bit): 22/22 unique, one-bit fraction 0.523, Hamming distance 49.8 %, no biased bit positions.
  • MAC (168 bit): 22/22 unique, one-bit fraction 0.497, Hamming distance 49.9 %, no biased bit positions.

The aggregate bit balance of the nonce and encrypted block deviates slightly from 0.5, but the deviation is not localised to any bit position, the 2014 sample leans the other way, and the pooled data are within or near chance. Further tests found no constant byte positions per key, no keystream reuse (first 8 encrypted bytes XOR nonce), no difference between same-key and cross-key MAC distances (50.2 % vs 49.6 %), and no values shared between 2014 and 2024.

Of 360 nonce-to-MAC and nonce-to-block byte correlations (exact Student's t p-values), only one survives Bonferroni correction (nonce[2] vs block[8], r = -0.78), and it does not replicate in the independent 2014 sample (r = +0.38), so it is most likely a false positive.

Statistical power: at n = 22 with Bonferroni correction, the smallest correlation detectable with 80 % power is |r| ≈ 0.77. No weakness is visible at this sample size, but this is a negative result with limited power, not evidence of strength.

Timing & Relay Feasibility

A complete exchange takes about 85 ms (83.8–87.4 ms across keys and cylinders), consistent with the vendor's stated release delay of approx. 0.1 s. The busy-poll period depends only on the key (1.1409 ms for key A, 1.1700 ms for key B), confirming that the key generates it.

The key does not impose a fixed response deadline. It polls until the cylinder is ready, and the number of polls for the same step varies between sessions (4–5 after identity, 7–8 after the encrypted block). A relay therefore does not have to answer within one bit slot: the fake cylinder holding the victim's key can answer "busy" until the real cylinder's reply arrives, hiding the forwarding latency inside the poll loop. The budget per step is at least one poll (~1.15 ms) and at most the key's unknown abort timeout.

Together with the absence of distance bounding, this makes a relay attack plausible. It has not been demonstrated; the decisive measurement is how many busy polls the key tolerates before aborting.

Identified Weaknesses

1. No Response Deadline or Distance Bounding (Severity: HIGH if the key's timeout is long)

Busy polling absorbs variable latency, so a relay between a legitimate key and a cylinder is plausible without breaking any cryptography.

2. Key Date & Time Sent in Plaintext (Severity: MEDIUM)

Every unlock reveals when each key is used, to the second, duplicating the audit trail.

3. System ID, Key Configuration & Cylinder Number in Plaintext (Severity: MEDIUM)

Identifies the installation, the key and the door.

4. Identity Format Fallback (Severity: UNKNOWN)

The key falls back from the 29-byte to the 27-byte identity format on error, a possible downgrade path.

5. Distinct Error Statuses (Severity: LOW)

0x21 after the encrypted block, 0x23 after the MAC and 0x34 for the identity format reveal which check failed.

6. Per-Key Timing & Clock Fingerprint (Severity: LOW)

The busy-poll period and clock offset differ per key, enabling passive key fingerprinting.

Limitations & Future Work

The dataset is 22 sessions from 2 keys, 2 cylinders and 1 installation, captured passively only (no chosen nonces, modified messages or fault injection), so the cipher and MAC constructions cannot be identified and generalisation to other CLIQ systems is unverified. Field semantics are inferred, though the timestamp and cylinder-number readings are backed by independent evidence.

Next steps:

  • Delay the cylinder's replies with a man-in-the-middle device and count how many busy polls the key tolerates, then build an instrumented two-device relay prototype.
  • Alter the timestamp in transit and check the cylinder's audit trail, to learn whether the key's time is authenticated.
  • Capture a cylinder that accepts the 29-byte identity to test the downgrade question.
  • Export the remaining Saleae recordings and capture more keys and installations.
  • Collect power or EM traces during the encrypted-block and MAC steps for side-channel analysis.

Summary

With a correctly framed decoder, the CLIQ key-cylinder protocol decodes cleanly into a five-step exchange with a simple additive checksum. Its cryptographic fields show no visible weakness at this sample size. The practical concerns are at the protocol level: the key's identity, date and time and the cylinder number travel in plaintext, and busy polling without a response deadline or distance bounding makes a relay attack plausible, though not yet demonstrated.