Modeling physical masquerade attacks

We have formally defined the security problem of a
stealthy adversary masquerading as a properly functioning agent.  perspectives from cybersecurity.

We introduced the concept of physical masquerade attacks in the context of multi-agent path finding, where a compromised insider in a multi-agent system attempts to gain access into unauthorized, forbidden locations without being noticed; we coined the term deviation attack for this type of malicious maneuver.

An illustration of the multi-agent path finding problem in the presence of deviation attacks, and of our defense approach based on co-observations.
An illustration of the multi-agent path finding problem in the presence of deviation attacks, and of our defense approach based on co-observations. The sensitive region is marked with S. Left: a traditional (minimal length) solution is not secure, since there exists a deviation attack for the square agent (dotted line).
Right: if we add the a constraint that agents need to observe each other in the middle of their paths, neither agent can reach the secure location without breaking with the expected observation schedule.

We provide a constraint-based formulation of multi-agent pathfinding that yields multi-agent plans that are provably resilient to physical masquerade attacks. This formalization leverages inter-agent observations to facilitate introspective monitoring to guarantee resilience. Furthermore, we have show that is important to distinguish between a cautious attacker (i.e., an attacker that aims to avoid detection under every possible circumstance) and a bold attackers (i.e., attackers that tolerate a risk of detection)

 

View all posts