Strahinja Janjusevic, a researcher at MIT’s Technology and Policy Program, is pioneering a new approach to maritime cybersecurity by blending deep learning with traditional control theory. As global trade relies heavily on massive merchant vessels, these legacy ships have become prime targets for sophisticated cyberattacks. The most pressing threat is GPS spoofing, where malicious actors transmit false signals to lure ships off course—a tactic already observed in contested international waters.

The proposed framework utilizes a sophisticated dual-layered defense. At its core, an internal LSTM (long short-term memory) autoencoder—a neural network that learns to compress and reconstruct data to identify patterns—monitors signal integrity to identify anomalies in incoming data. Simultaneously, a physics-based forecaster predicts the vessel’s movement by calculating environmental variables like wind and sea state. By cross-referencing these two outputs, the system can distinguish between harmless sensor noise and strategic cyberattacks, providing human operators with verified navigation data.

Beyond defensive measures, Janjusevic’s work extends to identifying emerging threats posed by AI agents. During research at Vectra AI, he explored how the Model Context Protocol (MCP)—the new standard for agent communication—could be weaponized for autonomous red teaming and command-and-control operations. This dual focus on technology and policy aims to bridge the gap between international national security efforts, ensuring that critical maritime infrastructure remains resilient against increasingly autonomous digital threats.

Strahinja Janjusevic, a researcher at MIT’s Technology and Policy Program, is pioneering a new approach to maritime cybersecurity by blending deep learning with traditional control theory. As global trade relies heavily on massive merchant vessels, these legacy ships have become prime targets for sophisticated cyberattacks. The most pressing threat is GPS spoofing, where malicious actors transmit false signals to lure ships off course—a tactic already observed in contested international waters.

The proposed framework utilizes a sophisticated dual-layered defense. At its core, an internal LSTM (long short-term memory) autoencoder—a neural network that learns to compress and reconstruct data to identify patterns—monitors signal integrity to identify anomalies in incoming data. Simultaneously, a physics-based forecaster predicts the vessel’s movement by calculating environmental variables like wind and sea state. By cross-referencing these two outputs, the system can distinguish between harmless sensor noise and strategic cyberattacks, providing human operators with verified navigation data.

Beyond defensive measures, Janjusevic’s work extends to identifying emerging threats posed by AI agents. During research at Vectra AI, he explored how the Model Context Protocol (MCP)—the new standard for agent communication—could be weaponized for autonomous red teaming and command-and-control operations. This dual focus on technology and policy aims to bridge the gap between international national security efforts, ensuring that critical maritime infrastructure remains resilient against increasingly autonomous digital threats.