Physics-Aware False Data Injection in Cyber-Physical Systems

This direction is merged into another direction: stealthy false data injection. Its last version stays readable and citable here.

Version 1 of 1222 papers353 conceptsfrozen 2026-10-04

Modern infrastructure—from power grids to autonomous vehicle fleets—relies on a "cyber-physical" loop. Sensors measure the physical world, a digital controller estimates the system's state, and actuators execute commands based on that estimate. For decades, the security of this loop rested on a comforting mathematical assumption: while sensors might fail or pick up noise, they won't all lie in a way that perfectly mimics reality.

Recent research has shattered this assumption, revealing a class of "stealthy" or "physics-aware" attacks that can blind operators and destabilize critical systems without triggering a single alarm.

The Blind Spot of State Estimation

To manage a complex system like a power grid, operators use a State Estimator. Because sensors are imperfect, the estimator looks at all incoming data and finds the "best fit" that obeys physical laws (like Kirchhoff’s laws). If a sensor reports a value that is mathematically impossible given the rest of the network, the system flags it as "Bad Data" and ignores it.

However, André M. H. Teixeira, Henrik Sandberg, and their colleagues demonstrated in A Cyber Security Study of a SCADA Energy Management System: Stealthy Deception Attacks on the State Estimator that an intelligent adversary can bypass this. By coordinating false data across multiple sensors, an attacker can inject an error that is mathematically consistent with the system's physics. To the controller, the grid appears to be in a valid—but incorrect—state. This allows an attacker to induce errors exceeding 50% of a line's capacity while remaining completely invisible to standard detection algorithms.

The Geometry of Stealth

Why does this work? In a linear system, the relationship between measurements ($z$) and the internal state ($x$) is defined by a matrix $H$, such that $z = Hx + e$ (where $e$ is noise). Standard detectors look at the "residual"—the difference between what the sensors say and what the model predicts.

A stealthy False Data Injection Attack (FDIA) works by choosing an attack vector $a$ that lies within the column space of $H$. As Deepjyoti Deka and Ross Baldick explain in Hidden Attacks on Power Grid: Optimal Attack Strategies and Mitigation, if the injected data perfectly fits the network's graph structure, the residual remains zero. The detector sees no "conflict" in the data, even though the underlying state is being manipulated.

This vulnerability isn't limited to power grids. In autonomous navigation, Tengchan Zeng and Aidin Ferdowsi note in Convergence of Communications, Control, and Machine Learning for Secure and Autonomous Vehicle Navigation that attackers can spoof sensor readings to steer vehicles off-course. Because the spoofed data mimics legitimate physical movement, the vehicle's navigation system accepts the lie as truth.

Moving Toward Active Defense

If the problem is that attackers know the system's "physics" too well, one solution is to change the physics.

Moving Target Defense (MTD) is a prominent strategy where operators subtly and randomly change the physical parameters of the system—such as the impedance of a transmission line using Distributed Flexible AC Transmission System (D-FACTS) devices. Mengxiang Liu, Jiming Chen, and others explore this in False Data Injection Attacks and Corresponding Countermeasure in DC Microgrid. By constantly shifting the "ground truth" of the network, they force the attacker to aim at a moving target, making it nearly impossible to craft a mathematically consistent (and thus stealthy) attack.

Another approach is Dynamic Watermarking. Matthew Porter and Ram Vasudevan describe this in Detecting Generalized Replay Attacks via Time-Varying Dynamic Watermarking as injecting a secret, low-power noise signal into the control inputs. If the sensor feedback doesn't contain the exact "echo" of this watermark, the system knows the data has been tampered with or replayed.

The Role of Machine Learning

As attacks become more sophisticated, defenders are turning to adversarial AI. Kejun Chen and Malik Hassanaly have proposed using Multi-Agent Reinforcement Learning (MARL) to create a "cat-and-mouse" training environment. In Adversarial Multi-Agent Reinforcement Learning for Proactive False Data Injection Detection, they show that by training a defender against an evolving AI attacker, the system can learn to spot the subtle statistical signatures of even the most "stealthy" injections.

Ultimately, the research direction across these 139 papers suggests that digital security (encryption and firewalls) is no longer enough. To protect the systems that run our world, we must reconcile the digital bits with the physical laws they represent, using the physics of the system as its own ultimate defense.

Papers in this version

Showing 60 of 222 papers.

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_physics_aware_false_data_injection_in_cy_25aed591,
  title = {Physics-Aware False Data Injection in Cyber-Physical Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/physics-aware-false-data-injection-in-cyber-physical-systems/rdr_25aed591559e4c21a148541c20c66fe4},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_physics_aware_false_data_injection_in_cy_25aed591_v1,
  title = {Physics-Aware False Data Injection in Cyber-Physical Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/physics-aware-false-data-injection-in-cyber-physical-systems/rdr_25aed591559e4c21a148541c20c66fe4/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}