Most industrial protocols still in service will accept a valid command from anyone who can reach the device on the network. Modbus, classic DNP3, IEC 60870-5-104, BACnet and the IEC 61850 multicast messages were designed for closed, trusted networks, so they carry no sender authentication and no integrity protection. Secure variants exist for nearly all of them, but adoption is slow because they need certificates, firmware support and engineering time on plants that run for decades. This guide goes protocol by protocol: what each one trusts, how attackers abuse that trust, which secure extension applies, and what to do on a plant where the secure version is not available. It is for OT security engineers, control system integrators and SOC analysts who already know the basics in the OT Security Guide and need the protocol detail.
Key takeaways
- Reachability equals authority for legacy protocols: a Modbus or DNP3 device treats any well-formed request as legitimate, so network access control is the first and often the only control.
- Every major protocol now has a secure form: Modbus/TCP Security, DNP3 Secure Authentication, IEC 62351 for IEC 104 and IEC 61850, CIP Security for EtherNet/IP, and the built-in OPC UA security model. Most are unused because devices, tools or certificates are missing.
- Secure protocols move the problem to identity: TLS and signed messages are only as good as the certificate issuance, trust lists, revocation and key storage behind them, which makes device certificates a non-human identity programme.
- Deep packet inspection beats port filtering: allowing TCP 502 says nothing about whether a host may write registers. Industrial firewalls and monitoring that understand function codes can restrict writes, programme downloads and mode changes.
- Real attacks use legitimate commands: FrostyGoop wrote Modbus holding registers and Industroyer2 spoke IEC 104. Detection must baseline who normally sends which commands, not look for malware signatures on the wire.
Why industrial protocols trust the network
Fieldbus and SCADA protocols were built for deterministic timing over serial links and isolated Ethernet segments. The design assumption was that only engineering workstations, HMIs and controllers could physically reach the wire. When these protocols were carried onto TCP/IP and later bridged towards business networks and remote access, the assumption broke but the protocol did not change. The result is a consistent set of weaknesses that the secure variants try to fix.
- No sender authentication: the device cannot tell an HMI from an attacker's laptop. Any host with a route can read or write. MITRE ATT&CK for ICS models this as T1692.001 Command Message under Impair Process Control (it replaced the revoked T0855 in ATT&CK v20).
- No integrity protection: values in transit can be altered by an adversary in the middle (T0830 Adversary-in-the-Middle), so an operator can see normal readings while the process drifts. Spoofed telemetry is modelled as T1692.002 Reporting Message.
- Engineering functions on the same port: programme upload and download, firmware update, operating mode change and diagnostics often share the port used for routine polling, so a read-only HMI path is also a reprogramming path (T0843 Program Download, T0858 Change Operating Mode).
- Rich discovery: device identification requests and tag browsing hand an attacker the vendor, firmware and point list needed to plan manipulation (T0888 Remote System Information Discovery, T0861 Point & Tag Identification).
- Fragile stacks: malformed or high-rate traffic can crash older controllers, so active scanning and some inspection techniques can cause a T0814 Denial of Service by accident.
Protocol reference: what each one trusts and its secure form
Ports below are the registered defaults. Vendors and integrators sometimes change them, so confirm against your asset inventory rather than relying on port numbers alone.
| Protocol | Where used | Default transport | Native security | Secure form |
|---|---|---|---|---|
| Modbus TCP | PLCs, drives, meters, building systems | TCP 502 | None: no authentication, no integrity, unit ID only | Modbus/TCP Security (TLS on TCP 802, role carried in an X.509 certificate extension) |
| Modbus RTU | Serial field devices, often behind gateways | RS-485 serial | None; CRC only detects line errors | No secure form; protect the gateway that bridges it to TCP |
| DNP3 | Electric, water and pipeline SCADA in North America and elsewhere | TCP or UDP 20000, serial | None in the base protocol | DNP3 Secure Authentication (SAv5 in IEEE 1815-2012, aligned with IEC 62351-5), TLS per IEC 62351-3 |
| IEC 60870-5-104 | European and international power telecontrol | TCP 2404 | None | IEC 62351-3 (TLS) and IEC 62351-5 (application-layer authentication) |
| IEC 61850 MMS | Substation client and server traffic | TCP 102 | Optional, rarely enabled | IEC 62351-4 (MMS security) over IEC 62351-3 TLS |
| IEC 61850 GOOSE and Sampled Values | Protection tripping and measurement on station and process bus | Layer 2 multicast, EtherType 0x88B8 (GOOSE) and 0x88BA (SV) | None | IEC 62351-6 message authentication; keys from IEC 62351-9 |
| OPC UA | Plant to MES, cloud and vendor integration | TCP 4840 (opc.tcp) | Built in: SecureChannel, certificates, user tokens | Native; must be configured (security mode Sign or SignAndEncrypt) |
| EtherNet/IP (CIP) | Discrete manufacturing controllers and I/O | TCP 44818, UDP 2222 for I/O | None in classic CIP | CIP Security: TLS for TCP, DTLS for UDP class 0/1 I/O |
| BACnet/IP | Building management systems | UDP 47808 | None in classic BACnet/IP | BACnet Secure Connect (BACnet/SC, TLS over WebSocket) |
Vendor-proprietary engineering protocols (for example Siemens S7comm on TCP 102 or the Unitronics PCOM service on TCP 20256) deserve the same treatment. The CISA advisory AA23-335A describes IRGC-affiliated actors reaching internet-exposed Unitronics PLCs on TCP 20256 with default or empty passwords, compromising at least 75 devices including at least 34 in the US water sector, replacing ladder logic and defacing HMIs.
Modbus: the protocol that trusts everyone
Modbus is a simple request and response protocol: a client sends a function code and an address range, and the server returns or writes register and coil values. There is no session and no identity beyond the unit identifier, which is a routing field, not a credential.
- Function codes matter more than ports: codes 1 to 4 read coils and registers; 5, 6, 15 and 16 write them; 8 runs diagnostics (some sub-functions restart communications); 43 returns device identification. A policy that allows reads from the historian and writes only from named HMIs removes most of the attack surface without touching the device.
- Register writes are the attack: the FrostyGoop malware (ATT&CK S1165) interacted with ENCO controllers over Modbus TCP 502, reading and writing holding registers. MITRE records its use in the January 2024 attack on a district heating company in Ukraine (C0041), where legitimate Modbus commands disrupted heating to civilians. Nothing in the traffic was malformed.
- Modbus/TCP Security: the Modbus Organization's secure variant wraps the protocol in mutually authenticated TLS on TCP 802 and lets the server authorise requests by a role carried in the client's X.509 certificate. Support is limited to newer devices, so most estates will run it only on new builds.
- Gateways are the weak joint: serial Modbus RTU devices are usually reached through a TCP to serial gateway. The gateway has no idea who sent the request, so restricting who can reach it is the only control for everything behind it.
DNP3 and IEC 60870-5-104 in utilities
DNP3 and IEC 104 carry telecontrol between control centres and outstations such as RTUs, reclosers and pumping stations. Both support unsolicited reporting, time synchronisation and control commands such as select-before-operate, direct operate and cold restart.
- DNP3 Secure Authentication: SAv5, specified in IEEE 1815-2012 and aligned with IEC 62351-5, adds an HMAC challenge and response so the outstation can confirm that critical requests (operate, restart, file transfer, configuration) come from a holder of the session key. It authenticates; it does not encrypt. Note that IEEE lists 1815-2012 as an inactive-reserved standard since March 2023, so check your vendor's implemented version and the current IEC 62351-5 edition when specifying new equipment.
- Key management is the hard part: SA uses update keys per user and outstation, and session keys derived from them. Without a process to distribute and rotate update keys, utilities ship devices with SA disabled or with a shared default key, which removes most of the benefit.
- IEC 104 is in the threat record: Industroyer (S0604), used against the Ukrainian grid in December 2016, carried IEC 101, IEC 104, IEC 61850 and OPC DA modules, and Industroyer2 (S1072) in 2022 spoke IEC 104 to high-voltage substations. Both relied on the protocol accepting commands from any reachable master.
- Compensating controls: restrict each outstation to its configured masters by IP, encrypt the WAN link (IPsec or private MPLS with encryption), and alert on control function codes from any source other than the SCADA master.
IEC 61850 and the substation
IEC 61850 splits substation traffic into client and server MMS (TCP 102) for SCADA and engineering, and multicast GOOSE and Sampled Values at Layer 2 for protection. GOOSE messages trip breakers and must arrive within a few milliseconds, which shapes what security is practical.
- GOOSE spoofing: any device on the station bus VLAN can publish a GOOSE frame with the right dataset reference and a higher state number. Subscribers accept it. IEC 62351-6 adds message authentication codes to GOOSE and SV; encryption is generally not used because of the timing budget.
- Key distribution: IEC 62351-9 covers certificate and key lifecycle for power systems, including enrolment protocols and a group key method for multicast. Without it, GOOSE authentication cannot be run at scale.
- Configuration files are sensitive: SCL files (SCD, ICD, CID) describe every logical node and dataset. An attacker with the SCD file knows exactly which GOOSE frame to forge. Protect them like source code.
- Segmentation inside the substation: separate station bus, process bus and engineering access, keep the VLANs for GOOSE tightly scoped, and disable unused switch ports. Layer 2 multicast cannot be filtered by an IP firewall.
OPC UA: secure by design, insecure by configuration
OPC UA (IEC 62541, published by the OPC Foundation as OPC 10000) has a built-in security model. The Part 2 Security Model, current release 1.05.06, describes a SecureChannel that uses keys agreed during the OpenSecureChannel exchange to sign and encrypt messages, with servers publishing their certificates through Discovery services. The weaknesses are in how deployments configure it.
- Security mode None: servers often expose an endpoint with security mode None and policy None for convenience during commissioning. Clients can then connect with no message protection. Disable it and allow only Sign or SignAndEncrypt.
- Deprecated policies: Basic128Rsa15 and Basic256 rely on SHA-1 and are deprecated by the OPC Foundation. Prefer Basic256Sha256, Aes128_Sha256_RsaOaep, Aes256_Sha256_RsaPss or the ECC policies in recent releases.
- Auto-trust of certificates: many servers ship with an option to accept any client application instance certificate, or engineers move rejected certificates into the trust list without checking them. That turns mutual authentication into a formality. Use a Global Discovery Server (GDS) or your own CA, and manage trust lists centrally.
- Anonymous user tokens: application authentication is separate from user authentication. Disable the Anonymous token, prefer X.509 or IssuedToken user identities over UserName, and map users to roles on the server.
- Stack vulnerabilities: OPC UA stacks have had denial of service and memory safety flaws. The OPC Foundation publishes advisories in its OPC-SecurityAdvisories repository; include the stack version in your inventory. INCONTROLLER (S1045) included OPC UA modules for discovering and interacting with servers.
Choosing a protection strategy
| Situation | Recommended approach | Trade-off |
|---|---|---|
| Legacy devices, no secure protocol support | Zone the devices, allow only named masters, add a protocol-aware firewall that blocks writes and programme functions except from engineering workstations | Insider or compromised HMI can still issue valid commands |
| Long WAN links to outstations | IPsec or encrypted private WAN plus source restriction; DNP3 SA or IEC 62351-5 where outstations support it | Encryption adds latency and key management at many remote sites |
| New greenfield line or substation | Specify secure variants (CIP Security, Modbus/TCP Security, IEC 62351, OPC UA SignAndEncrypt) in procurement, with certificate enrolment | Requires an OT PKI and staff who can run it |
| IT to OT data flows (MES, cloud analytics) | OPC UA through the Level 3.5 DMZ, server pushes or a data diode where feasible, no inbound writes | Diodes remove bidirectional use cases |
| Small site with limited staff | Remove internet exposure, change default passwords, restrict to one engineering workstation, back up logic, use a managed monitoring service | Visibility depends on the provider |
Large operators with many sites should standardise on a small number of secure profiles and build the OT PKI once. Smaller organisations get more risk reduction from removing exposure and enforcing function-code allowlists than from a partial secure-protocol rollout. Procurement is the long-term lever: see the OT Product Procurement Guide.
Where identity fits
Every secure industrial protocol replaces "trusted because reachable" with "trusted because it holds a key". That makes PLCs, RTUs, IEDs, OPC UA servers and gateways machine identities with certificates, symmetric keys and service credentials that must be issued, stored, rotated and revoked.
- Device certificates: application instance certificates in OPC UA and device certificates in CIP Security and IEC 62351-3 need an issuing CA, a validity period that matches maintenance windows, and a revocation path that works offline. See the Device and IoT Identity Guide.
- Shared keys: DNP3 SA update keys and GOOSE group keys are shared secrets; record where each is held and who can change it.
- Engineering credentials: project passwords, PLC access passwords and vendor default accounts are the real gate to programme download. Treat them as privileged accounts. See the OT and ICS Identity and Access Guide.
Detection and measurement
Passive monitoring on SPAN or TAP ports is the safe default in OT, because it does not inject traffic into fragile stacks. Protocol-aware sensors should parse function codes and build a baseline of who talks to whom.
- Alert on: Modbus write codes (5, 6, 15, 16) or diagnostics (8) from hosts that are not approved HMIs; DNP3 or IEC 104 control commands from a non-master source; programme download, firmware update or stop and run mode changes outside a change window; a new OPC UA client certificate or a session using security mode None; GOOSE frames from an unexpected source MAC or with sudden state number jumps; Modbus device identification sweeps (T0846, T0888).
- Log at the edge: industrial firewalls should log denied writes and engineering functions, and those logs should reach the SOC. See the OT Network Security Monitoring Guide.
Useful metrics:
- Percentage of controllers and RTUs reachable only from their documented masters (target 100%).
- Number of OT devices reachable from the internet (target zero; check with external attack surface scans).
- Percentage of OPC UA endpoints with security mode None disabled.
- Percentage of write and programme-download traffic coming from approved engineering hosts, as seen by monitoring.
- Percentage of new equipment purchased with a secure protocol profile enabled at commissioning.
- Device certificates expiring within 30 days without a scheduled renewal.
Implementation approach
- Inventory flows: capture passively for at least one full production cycle and record each protocol, source, destination and function code. See the OT Asset Visibility Guide.
- Remove exposure: close any internet-reachable control ports and replace direct vendor connections with brokered remote access. See the OT Remote Access Guide.
- Zone and restrict: place devices into zones and allow only documented conduits. See the OT Network Segmentation, Purdue Model and IEC 62443 Zones and Conduits Guide.
- Function-code rules in monitor mode: deploy protocol-aware firewall rules that log, not block, for a full cycle; review every would-be block with operations.
- Enforce in rings: enforce on a non-critical cell first, then by area during planned outages, with a tested bypass (a pre-approved rule set to restore) as the rollback.
- Enable secure variants where supported: start with OPC UA (configuration only), then CIP Security or IEC 62351 on new or refurbished equipment, after the OT PKI and key processes are in place.
- Test on a bench: validate latency, failover and certificate expiry behaviour on a test rig before enabling authentication on protection or safety-related traffic.
Common failure patterns
- Port-only firewall rules: allowing TCP 502 or 44818 between zones permits writes and reprogramming as well as reads.
- Default and empty passwords on engineering services: the Unitronics campaign in AA23-335A needed nothing more. Default credentials are now ATT&CK for ICS T1694.001.
- Secure mode enabled, trust disabled: OPC UA servers set to auto-accept certificates, or DNP3 SA with a shared factory key, look compliant and give no assurance.
- Certificates with no renewal plan: a device certificate expiring on a plant that cannot be restarted until the next outage causes the availability incident that security was meant to prevent.
- Active scanning of fragile devices: uncoordinated vulnerability scans can crash older controllers; use passive discovery and vendor-approved queries.
- Flat VPN into OT: a remote access account that lands on a flat control network bypasses every protocol control. The Colonial Pipeline ransomware attack began with a single unused VPN account, and the operator shut down pipeline operations as a precaution.
Framework mapping
| Framework | Reference | How it applies |
|---|---|---|
| ATT&CK for ICS mitigations | M0802 Communication Authenticity, M0937 Filter Network Traffic, M0807 Network Allowlists, M0930 Network Segmentation | Secure protocol variants, function-code filtering, master allowlists, zoning |
| NIST SP 800-53 Rev. 5 | SC-8, SC-23, IA-3, AC-4, SC-7, SI-4 | Transmission integrity, session authenticity, device authentication, flow enforcement, boundary protection, monitoring |
| NIST CSF 2.0 | PR.DS-02, PR.IR-01, DE.CM-01 | Data in transit, network protection, network monitoring |
| ISA/IEC 62443-3-3 | SR 1.2, SR 3.1, SR 4.1, SR 5.1 | Device authentication, communication integrity, confidentiality, network segmentation |
| ISO/IEC 27001:2022 Annex A | A.8.20, A.8.21, A.8.22 | Network security, security of network services, segregation of networks |
NIST SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security, remains the final version; NIST has released an initial public draft of Rev. 4 for comment, so check for changes before citing section numbers. See the IEC 62443-3-3 and 4-2 System and Component Security Levels Guide.
Practitioner checklist
- Map every industrial protocol flow with source, destination and function codes from passive capture.
- Confirm no controller, RTU, HMI or gateway is reachable from the internet, and change all default and empty passwords.
- Restrict each device to its documented masters and engineering workstations.
- Deploy protocol-aware rules that allow reads broadly and writes, downloads and mode changes narrowly; test in monitor mode first.
- Disable OPC UA security mode None, deprecated policies and Anonymous user tokens; manage trust lists centrally.
- Specify secure protocol profiles (CIP Security, Modbus/TCP Security, IEC 62351, DNP3 SA) in procurement, and build the OT PKI and key processes to support them.
- Protect SCL files, PLC projects and engineering passwords as privileged assets.
- Alert on control and engineering commands from unexpected sources and outside change windows.
- Track device certificate expiry and rotate shared protocol keys on a schedule aligned to outages.
Standards and references
- NIST: SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security
- OPC Foundation: OPC 10000-2 OPC UA Part 2 Security Model
- ODVA: CIP Security
- IEEE: 1815-2012 Standard for Electric Power Systems Communications, DNP3
- CISA: AA23-335A IRGC-Affiliated Cyber Actors Exploit PLCs in Multiple Sectors
- OPC Foundation: Security
- MITRE ATT&CK: FrostyGoop
- MITRE ATT&CK: Command Message
Related NHI Mgmt Group resources: OT Security Guide · OT Network Security Monitoring Guide · OT and ICS Identity and Access Guide · IEC 62443 Guide