CAN (Controller Area Network) is the broadcast serial bus Bosch released in 1986 and ISO standardized as ISO 11898 — every node sees every message, any node can transmit, and the protocol has no field for a sender identity, a credential or a message signature anywhere in its design. That design choice shows up directly in two real CISA ICS alerts, and it has not changed through CAN FD; only the newest generation, CAN XL, has an optional (and still maturing) security layer called CANsec. For machine builders putting CAN-connected controllers into industrial equipment, vehicles or off-highway machinery, the practical question is less "is CAN secure" and more "what, if anything, sits on top of CAN to compensate for the fact that it isn't."
Key takeaways
- Robert Bosch GmbH introduced CAN at the SAE congress in February 1986 to cut the point-to-point wiring of automotive electronics.
- Four attack patterns follow directly from the table above.
- Neither alert resulted in a CVE identifier (both predate the routine CVE-tagging now standard on most CISA ICS advisories), and neither is remotely exploitable over a network.
- AUTOSAR's Secure Onboard Communication protocol specification (document 969, release R24-11) adds an Authenticator, normally a MAC, and optionally a Freshness Value to each protected message.
What is CAN, and why was it never designed with security in mind?
Robert Bosch GmbH introduced CAN at the SAE congress in February 1986 to cut the point-to-point wiring of automotive electronics. Per CAN in Automation's history of CAN technology, Bosch's CAN 2.0 specification went to ISO in the early 1990s, ISO 11898 was published in November 1993, and a 1995 addendum added the extended frame format with 29-bit identifiers. ISO splits the standard into parts: ISO 11898-1 specifies the data link layer, and ISO 11898-2 the high-speed physical layer. CAN's defining property is that it is a multi-master, broadcast bus: when the bus is idle, any connected node may transmit, and every other node on the same segment receives and can act on every frame, with priority resolved by non-destructive bitwise arbitration rather than a central controller. That design, optimized for deterministic low latency and minimal per-node silicon cost inside a single trusted wiring harness, is exactly why CAN has no sender-authentication field: the protocol was never meant to distinguish a legitimate electronic control unit (ECU) from any other device wired to the same bus. CAN today runs far beyond passenger cars: SAE J1939 standardizes its use in trucks, buses, construction and agricultural machinery, and industrial automation uses CANopen for motor drives, I/O modules and other machine-control devices.
| CAN design property | Security implication |
|---|---|
| Broadcast, multi-master | Every node reads every frame; there is no way to address a message privately to one recipient at the protocol level |
| No sender-identity field | A receiving node cannot verify which physical device actually sent a given frame |
| No authentication or encryption in the base frame | Any device electrically connected to the bus can inject, by design, as valid a frame as any legitimate ECU |
| Priority-based arbitration, no central access control | Nothing in the protocol itself limits which node may transmit or when, beyond message priority |
How do bus-off and frame-injection attacks work?
Four attack patterns follow directly from the table above. The Fraunhofer IPMS CANsec whitepaper lists the same set in its threat model:
- Injection and spoofing. A node sends a frame with the identifier a legitimate ECU uses. Receivers act on it because nothing in the frame proves its origin.
- Replay. A recorded frame, such as an unlock command, is sent again later. Classic CAN has no freshness field to reject it.
- Arbitration flooding. A node that keeps sending identifier 0 always wins arbitration, so lower-priority traffic never gets through.
- Bus-off. CAN's fault confinement is turned against a victim. Every controller keeps transmit and receive error counters. An attacker who overwrites selected bits of the victim's frames makes the victim see its own transmit errors, and the counter climbs until the controller removes itself from the bus.
The thresholds, as the Linux kernel encodes them in its SocketCAN state definitions:
| Controller state | Error count | What the node can still do |
|---|---|---|
| Error-active | below 96 | Transmit and signal errors normally |
| Error-warning | 96 to 127 | Same, with a warning flag raised |
| Error-passive | 128 to 255 | Transmit, with restricted error signalling |
| Bus-off | 256 or more | Nothing until restarted |
Recovery is a configuration choice. In the SocketCAN documentation, an automatic restart after bus-off happens only if restart-ms is set, and "by default it's off". A bus-off attack against a node with no restart policy silences it until a reset.
What have real CISA alerts found in CAN bus deployments?
Two CISA alerts, years apart, both requiring physical or electrical access to the bus rather than a remote network attack — consistent with CAN's design as a bus for a physically enclosed wiring harness, not a networked, internet-facing protocol.
| Alert | Year | What it found | Access required | CISA's mitigation |
|---|---|---|---|---|
| ICS-ALERT-17-209-01 | 2017 (last revised) | A bit-level manipulation technique (targeting individual recessive/dominant bits rather than whole frames) that forces a CAN node into its "bus-off" error state, silencing it — disclosed by researchers Andrea Palanca, Eric Evenchick, Federico Maggi and Stefano Zanero | Electrical connection to the bus, typically via an automobile's OBD-II port | "Limit access to input ports (specifically OBD-II) on automobiles" |
| ICS-ALERT-19-211-01 | 2019 | Insecure CAN bus implementation in small-aircraft avionics, reported by Rapid7: an attacker could inject false engine, compass/attitude, altitude, airspeed or angle-of-attack readings indistinguishable from real ones by a pilot reading cockpit instruments | Unsupervised physical access to the aircraft, to attach a device to the avionics CAN bus | Restrict physical aircraft access; manufacturers evaluate CAN bus filtering, whitelisting and network segregation |
Neither alert resulted in a CVE identifier (both predate the routine CVE-tagging now standard on most CISA ICS advisories), and neither is remotely exploitable over a network. The common thread: with no sender authentication, whoever can physically or electrically reach the bus can act as any other node on it, whether that bus sits in a car, a small aircraft or an industrial machine's control cabinet.
What does AUTOSAR SecOC add to a CAN frame?
AUTOSAR's Secure Onboard Communication protocol specification (document 969, release R24-11) adds an Authenticator, normally a MAC, and optionally a Freshness Value to each protected message. Both sender and receiver must run a SecOC module, and the receiver must know the Freshness Value the sender used. The specification assumes mainly symmetric MACs, because they need much smaller keys than signatures, but allows asymmetric schemes. On classic CAN, with only 8 data bytes, the MAC and freshness value are truncated. The three predefined profiles all use AES-128 CMAC per NIST SP 800-38B:
| Profile | Truncated MAC | Truncated freshness | Note in the specification |
|---|---|---|---|
| 1 (24Bit-CMAC-8Bit-FV) | 24 bits | 8 bits | Least significant 8 bits of the freshness value sent |
| 2 (24Bit-CMAC-No-FV) | 24 bits | none | Only if no synchronized freshness value is established |
| 3 (JASPAR) | 28 bits | 4 bits | Counter-based freshness with master-slave synchronization; "shall be used for CAN" |
SecOC is a design decision per message: someone has to choose which signals are authenticated, provision keys to each ECU and keep freshness counters synchronized. Fraunhofer notes that because SecOC runs in software above the data link layer, it can cost significant host CPU time.
Do CAN FD or CAN XL fix CAN's lack of security?
Not fully, and only the newest generation attempts to. CAN FD (Flexible Data Rate), which Bosch introduced in 2012 and ISO later standardized in ISO 11898-1 after adding safeguards such as a stuff-bit counter, raises the data rate and the payload per frame (from 8 to 64 bytes) but keeps the broadcast, no-authentication model; it is a throughput upgrade, not a security one. CAN XL, standardized in ISO 11898-1:2024, is the first generation with an optional security mechanism. Per CiA's CAN XL page, CAN XL carries 1 to 2048 data bytes, splits the old identifier into an 11-bit priority ID and a 32-bit acceptance field, and reaches 20 Mbit/s and higher with PWM coding. CANsec is specified in CiA 613-2, "CANsec data plane": version 1.0.0, dated 29 May 2026, has Draft Standard Proposal (DSP) status. Its scope reuses "the security functionalities of MACsec" and leaves "the specifics of a key exchange/distribution protocol out of scope."
| AUTOSAR SecOC | CANsec (CiA 613-2) | |
|---|---|---|
| Layer | Above the data link layer, per PDU | Data link layer, per CAN XL frame |
| Bus generations | Classic CAN, CAN FD, others | CAN XL only |
| Overhead | Truncated MAC + freshness (32 bits in Profiles 1 and 3) | 4-byte header + 16-byte authentication tag |
| Encryption | No (authenticity and freshness) | Optional |
| Typical implementation | Software in the ECU | Controller hardware or IP core (AES-GCM in Fraunhofer's proof of concept) |
| Key distribution | Left to the system design | Out of the specification's scope |
A device described as "running CAN XL" is not automatically running CANsec; treat them as two claims to verify against product documentation.
What does J1939 change for trucks and off-highway machines?
J1939 adds addressing on top of CAN but no authentication. Per the Linux kernel J1939 documentation, the 29-bit identifier holds a 3-bit priority, an 18-bit Parameter Group Number (PGN) and an 8-bit source address. Every ECU is known by a 64-bit NAME and takes its source address through address claiming (J1939-81). Both the source address and the NAME are values the sender writes itself, so a device on the bus can send parameter groups under another ECU's address or claim an address it does not own. Transport protocols carry up to 1,785 bytes over multiple frames, so a spoofed multi-frame message is as easy as a single-frame one. J1939 security reviews therefore come back to bus access, gateways and firmware, as for any CAN network.
Worked example: sizing SecOC for a classic CAN brake signal
Synthetic example, for illustration only. An off-highway machine sends a 16-bit brake-pressure value, a 4-bit status and a 4-bit rolling counter in one 8-byte classic CAN frame at 500 kbit/s.
Count free bytesThe signals use 3 bytes. SecOC Profile 1 needs 4 bytes (24-bit MAC + 8-bit freshness), so 7 of 8 bytes are used and 1 is spare. Reject Profile 2With no freshness value, a recorded "brake released" frame replays cleanly. Estimate forgery costA random 24-bit MAC guess succeeds once in 2^24 (about 16.8 million) tries. An 8-byte standard frame is 111 bits before stuff bits, so a 500 kbit/s bus carries at most about 4,500 such frames per second. 2^24 tries take about 62 minutes of a fully saturated bus, which a gateway or intrusion monitor should notice long before. Plan freshness recoveryOnly 8 bits of the counter are sent, so the receiver rebuilds the full value; after a reset or a bus-off, both ECUs need a resynchronization path.
What are the common pitfalls?
- Authenticating messages but not protecting availability. SecOC and CANsec do not stop arbitration flooding or bus-off attacks.
- Leaving bus-off recovery undefined. A controller with no restart policy stays silent after an attack.
- One key for every ECU. Extracting the key from one cheap node lets it forge messages for the whole bus.
- Trusting the gateway's own firmware blindly. The gateway that separates domains is the most valuable target on the network.
- Treating "CAN XL" as "CANsec". The security add-on is optional and its key distribution is left to the integrator.
What should a machine builder or integrator check first?
Inventory every CAN-connected ECU and controller, and every physical access point to the bus — OBD-II or diagnostic connectors, harness junctions, service ports — since CAN's security model depends on controlling who can reach the bus. Diagnostic access deserves its own review; see UDS protocol security. Treat bus access as the actual security boundaryRestrict and monitor diagnostic ports and harness access points, as ICS-ALERT-17-209-01's mitigation implies, rather than assuming a locked enclosure is enough. Segment safety- or mission-critical CAN domainsfrom telematics, infotainment and aftermarket tools behind a gateway ECU that filters what crosses between them. Check whether message authentication is actually designed in— SecOC on classic CAN or CAN FD, CANsec on CAN XL — and how keys are provisioned and rotated. For road vehicles this belongs in the ISO/SAE 21434 threat analysis. Fold CAN-connected controller firmware into the same SBOM and firmware risk review as any other embedded deviceThe ECU firmware, not the bus, is usually where outdated stacks and other exploitable weaknesses live; the firmware SBOM guide covers how to inventory it. For plant-floor CAN and CANopen networks, the NIST SP 800-82 OT guide covers segmentation and monitoring.
What can AI do in a CAN-connected ECU's firmware scan?
CAN's lack of authentication means the real security work sits in the ECU firmware and in whatever authentication layer the builder designed in. SoftWSP's pipeline extracts the software components from a CAN-connected controller's firmware image into a CycloneDX 1.6 SBOM and matches them against NVD, the CISA KEV catalog, FIRST EPSS and CISA CSAF ICS advisories, the same process used on any embedded device. AI drafts a plain-language report that cites the source for each finding — which component, which CVE, whether it is KEV-listed — for an engineer to review. The scan does not test the bus electrically, does not detect whether SecOC or CANsec is configured, and does not confirm that message authentication is enforced at runtime; those are protocol-level tests an engineer still has to run. The free SBOM viewer opens the resulting component list.
Knowing which components and versions a CAN-connected controller's firmware is built from, and which of them carry known exploited vulnerabilities, is firmware-level work that a bus specification cannot answer: see SoftWSP's device & firmware security services.
This guide draws on ISO's catalogue for ISO 11898-1, two CISA ICS alerts, AUTOSAR's published SecOC protocol specification, CiA's CAN XL and CiA 613 pages, the Linux kernel's SocketCAN and J1939 documentation, SAE International's J1939 top-level document description and a Fraunhofer IPMS whitepaper on CANsec. The full CiA 613-2 text is available to CiA members, so CANsec details here follow CiA's public summary; the current CiA specification and a specific product's documentation are the authoritative sources for exact fields. This guide is general information, not a substitute for the purchased ISO/SAE/CiA standard texts, a vendor's product-specific security documentation, or a current CISA ICS advisory search for your exact hardware.
Put this guide to work
Frequently asked questions
Does the CAN bus have any built-in authentication or encryption?
No. CAN (Controller Area Network) is a broadcast, multi-master serial bus: per its design, any node connected to the bus can transmit when the bus is idle, and every other node receives and processes every frame — there is no concept of a sender identity field, a credential, or a message-origin check anywhere in the base protocol. Bosch released CAN publicly in 1986 for automotive wiring-harness reduction, and it became an ISO standard (ISO 11898) in 1993; the standard has been revised several times since — most recently as a three-edition structure (ISO 11898-1:2024, covering the data link layer and physical coding sublayer, superseding the 2015 edition) — but no revision has added authentication, encryption or integrity-checking to the base CAN frame itself. This is a direct consequence of CAN's original design goals: minimal per-node hardware cost and deterministic, low-latency arbitration between electronic control units (ECUs) in a trusted, physically enclosed wiring harness, not resistance to an attacker who can already reach the bus electrically.
What have real CISA alerts found in CAN bus deployments?
Two, both requiring physical or electrical access to the bus rather than remote network access, and neither resulting in a CVE assignment (both predate CISA's routine CVE tagging of every ICS finding). CISA's ICS-ALERT-17-209-01 (last revised 3 August 2017) describes a denial-of-service technique researchers Andrea Palanca, Eric Evenchick, Federico Maggi and Stefano Zanero disclosed through coordinated vulnerability disclosure: rather than targeting whole frames, the technique manipulates individual recessive/dominant bits to force a CAN node into its error-handling "bus-off" state, silencing it. CISA's stated mitigation was narrow — "limit access to input ports (specifically OBD-II) on automobiles" — reflecting that the underlying weakness is architectural to CAN itself, not a single vendor's bug to patch. CISA's ICS-ALERT-19-211-01 (30 July 2019), based on a report from Rapid7, covers insecure CAN bus implementations in small-aircraft avionics: an attacker with unsupervised physical access to the aircraft could attach a device to the avionics CAN bus and inject false engine telemetry, compass/attitude, altitude, airspeed or angle-of-attack data that a pilot reading cockpit instruments would have no way to distinguish from legitimate readings. CISA's recommended mitigations were physical-access restriction plus manufacturer-level CAN bus filtering, whitelisting and segregation — the same category of control the automotive industry had already begun adopting for the 2017 finding.
Do CAN FD and CAN XL fix CAN's lack of security?
CAN FD (Flexible Data Rate, introduced by Bosch in 2012 and later standardized in ISO 11898-1) increases maximum data rate and payload size per frame but adds no authentication, encryption or integrity mechanism — it inherits the same broadcast, no-sender-identity design as classic CAN. CAN XL, standardized in the current ISO 11898-1:2024 (Edition 3, which specifies three protocol generations — CAN CC/classic, CAN FD and CAN XL — in one 82-page document replacing the 2015 edition), is the first generation where an optional security layer exists: CANsec, documented as CiA 613-2 by CAN in Automation, is explicitly modeled on IEEE 802.1AE MACsec (the Ethernet security standard) and provides authenticated, optionally encrypted CAN XL frames using AEAD. As of November 2026, CANsec is a working specification under active development by a CiA task force, not a settled, broadly deployed standard — independent published research describes ongoing work on cipher-suite selection and on reusing existing MACsec implementations as CANsec's security engine, which is a sign of an evolving rather than finished ecosystem. A buyer or integrator should treat "the device uses CAN XL" and "the device has CANsec enabled and configured" as two separate claims to verify, not one.
Is there anything already deployed today that authenticates CAN messages?
Yes, but it sits above CAN itself, not inside it, and it is not universal. AUTOSAR — the automotive software architecture standard most passenger-vehicle and many commercial-vehicle ECUs are built on — defines SecOC (Secure Onboard Communication), a module that adds a message authentication code and freshness counter to selected CAN messages at the application layer, using a symmetric key shared between sender and receiver ECUs. SecOC has to be deliberately designed into a vehicle's or machine's electrical architecture message-by-message (which signals get authenticated, which don't, and how keys are provisioned and rotated) — it is not a property CAN itself provides, and a machine builder cannot assume a CAN-based control system has it just because the ECUs happen to run AUTOSAR. Industrial equipment built on SAE J1939 — the higher-layer protocol standardizing CAN use in trucks, construction equipment, agricultural machinery and other off-highway vehicles via extended 29-bit CAN identifiers — has no equivalent built-in authentication layer of its own; published security research on J1939 describes the same broadcast trust model and resulting attack surface as classic automotive CAN, just applied to a different fleet of machines.
Sources
- ISO: ISO 11898-1:2024 — Road vehicles — Controller area network (CAN) — Part 1: Data link layer and physical coding sublayer (Edition 3, published 2024-05, 82 pages, supersedes the 2015 edition)
- ISO: ISO 11898-1:2015 — Road vehicles — Controller area network (CAN) — Part 1: Data link layer and physical signalling (withdrawn edition, scope record)
- CISA ICS Alert ICS-ALERT-17-209-01: CAN Bus Standard Vulnerability (last revised 3 August 2017)
- CISA ICS Alert ICS-ALERT-19-211-01: CAN Bus Network Implementation in Avionics (30 July 2019)
- SAE International: J1939_202306, Serial Control and Communications Heavy-Duty Vehicle Network — Top-Level Document
- Fraunhofer IPMS: CANsec — Security for the Third Generation of the CAN Bus (whitepaper PDF)
- AUTOSAR: Specification of Secure Onboard Communication Protocol (FO R24-11, document 969) — Profiles 1-3
- CAN in Automation (CiA): CAN XL — data link layer, CANsec add-on summary
- CAN in Automation (CiA): CiA 613 series, CAN XL add-on services (CiA 613-2 v1.0.0 CANsec data plane, DSP, 2026-05-29)
- CAN in Automation (CiA): History of CAN technology
- Linux kernel: SocketCAN documentation (controller states, restart-ms)
- Linux kernel source: include/uapi/linux/can/netlink.h (CAN controller state and error-count thresholds)
- Linux kernel: J1939 documentation (29-bit identifier layout, NAME, address claiming, transport protocol)
About this guide
SoftWSP OT/ICS Desk
Covers the OT/ICS security cluster: 87 guides.

In this cluster
Questions about this page?
