PLC remote attestation is difficult in many industrial control systems because programmable logic controllers (PLCs) often lack trusted hardware that can authenticate a software-integrity report. A checksum or hash may measure control logic, but without a protected key or hardware root of trust, a verifier has limited ways to confirm that the measurement is genuine. Physics-based attestation, known as PAtt, offers a different approach: it connects the logic measurement to the physical behavior of the process.
Rather than relying only on a message sent over the network, PAtt encodes information about the PLC’s memory state into actuator commands. The resulting sensor readings provide evidence that the expected control logic and physical system produced the response. The approach was implemented on a PLC-controlled robotic arm and evaluated against specific attacks; its results demonstrate feasibility, not a guarantee for every PLC or process.
Why PLC integrity is hard to verify
A PLC bridges supervisory systems and field-level equipment. It reads sensor inputs, executes control logic, and sends outputs to actuators, often in recurring scan cycles. If an attacker changes the logic, the controller may alter how the process is operated while attempting to conceal the change from a supervisory system.
Legacy and existing PLCs may not include a hardware root of trust, such as a trusted execution component, to protect and authenticate measurements. Some devices can calculate hashes over control-logic memory, but that capability alone does not prove to a remote verifier that the reported hash is authentic. Coverage can also be limited: in the reported implementation, hashing covered logic-accessible memory, not the real-time operating system.
Conventional software-based attestation can rely on strict assumptions about execution timing and communication. PAtt adds a physical check: the verifier evaluates sensor traces that should correspond to the reported measurement and the expected process.
How physics-based PLC remote attestation works
The verifier begins by sending the PLC a fresh random challenge, or nonce. The PLC makes that value available to its control logic and calculates a cryptographic hash over selected logic objects and the nonce. Because each challenge is fresh, an earlier response should not simply serve as the current attestation.
The PLC then maps the hash to an actuation strategy. In the robotic-arm example, the strategy determines the order of small movements used to reach the process’s normal goal. Different movement sequences can reach the same endpoint while producing different sensor traces along the way. The strategy is designed to stay within safety constraints and avoid disrupting the normal process objective.
As the actuator follows the selected path, sensors record the system’s response. The PLC sends the measurement information and sensor readings to the verifier. The verifier checks the hash against the expected logic and nonce, then determines whether the trace is consistent with the hash and the physical process.
From movement traces to an integrity decision
PAtt uses sensor data to reconstruct the small actuator commands encoded by the hash. In the reported implementation, a machine-learning classifier was trained on normal-operation data, using features such as position and statistical properties of accelerometer and gyroscope signals. The verifier compares the reconstructed result with the expected hash using a weighted Hamming distance. Classifier confidence weights the comparison, and a threshold determines whether to raise an alarm.
This combines two checks: whether the software measurement is correct, and whether the associated physical response appears consistent with that measurement. The physical process contributes system-specific variation to the readings, making them more informative than a hash value sent on its own.
What the evaluation demonstrated
The researchers implemented PAtt on a Siemens S7-1200 PLC controlling a robotic arm. The testbed used inertial measurement units on the arm segments to collect sensor readings. Because the PLC could execute only a limited number of micro-commands in each scan cycle, the 256-bit hash-derived strategy was divided across multiple rounds. The implementation generated eight micro-commands per scan cycle and used fresh nonces across 32 rounds.
The evaluation considered hash-approximation attacks that flipped one or two bits, as well as attempts to replay or reconstruct sensor readings. The detector was trained using normal-operation data rather than prior examples of attacks. With a threshold selected to balance detection and false alarms, the reported results achieved 97% accuracy against the tested attacks. A stricter threshold reduced false positives during normal operation but also lowered detection performance for one-bit changes.
That trade-off matters operationally. Detection performance depends on the chosen threshold and the process data used to model normal behavior. The reported accuracy applies to the tested setup and attack cases; it should not be treated as a universal performance figure or assumed to transfer unchanged to another production environment.
Practical considerations for industrial deployment
PAtt is intended for systems where the control process can safely perform alternative actuation sequences that reach the required operating goal. An attestation routine must not interfere with safety constraints or normal production. Some processes have critical zones—for example, when equipment is engaged in a fixed operation—that restrict when attestation can run. Systems with no suitable opportunity for safe, bounded movement may not be appropriate for this method.
Deployment also requires a usable model of normal sensor behavior and a way to collect reliable traces. PLC capabilities differ, and memory-measurement support may depend on available APIs or require engineering work in the control logic. Network protections such as TLS can help protect communications, but they do not replace the physical-response checks or resolve limits in what memory the PLC can measure.
Conclusion
Physics-based attestation extends PLC integrity checking by tying a software measurement to actuator behavior and sensor evidence. For suitable processes, it offers a way to detect certain control-logic manipulations and replay attempts without adding a traditional hardware trust anchor. Its practical value depends on safe actuation opportunities, representative normal-operation data, measurement coverage, and careful threshold selection. Engineers considering the approach should evaluate those constraints against the safety and operating requirements of the specific control system.