Abstract: Gaming has always been a very prevalent part of our society. There is a long history of using various board games as methods of covert communication, from what we could call "differential" communication (i.e. the movement of pieces constitutes the "language") to the idea of "game state" communication (where the current state of all the manipulatable objects in the game world represents a message). For many years we have seen covert channels in games like chess, checkers and other simple board games and even evolution to more "complex" computer games like magnetron. Enter the 21st century and the advent of the high performance personal computer.
Introduction
This research explores the use of "sandbox" video games—specifically Garry’s Mod—as a medium for covert communication. By manipulating the "physical" properties of in-game objects (props), the authors demonstrate how a digital world can host a "physical" covert channel that is difficult to distinguish from normal, albeit eccentric, gameplay.
Problem Statement
Covert channels are communication paths not intended for information transfer, often used to exfiltrate data or communicate secretly. Traditionally, these are purely digital (e.g., hiding data in TCP/IP headers) or purely physical (e.g., a dead drop under a bridge).
The authors identify a gap in the research regarding game-based covert channels. While previous work looked at minute player movements in shooters like Quake 3, this paper tackles the potential of "sandbox" environments. In these games, the "game state"—the arrangement, color, and properties of manipulatable objects—can be used to encode complex messages. The challenge is creating a channel that is robust enough to handle "noise" from other players while remaining inconspicuous to server administrators.
Methodology
The authors developed a proof-of-concept (PoC) using Garry’s Mod, a game built on Valve’s Source Engine that allows players to spawn and modify objects.
- The Encoding Scheme: The authors used 55-gallon drum "props" as data containers. Each prop has RGBA (Red, Green, Blue, Alpha) color values. They encoded one ASCII character into each of the R, G, and B channels (0-255). Thus, a single barrel carries three characters of data.
- The "Key" (Filtering): To separate the message from "noise" (barrels spawned by other players), they used the Alpha (transparency) channel. By setting the Alpha to a specific, non-standard value (e.g., a random number not ending in 0 or 5), the receiver’s script can identify which objects belong to the covert message.
- Message Ordering: Data packets were ordered based on the sequence in which the props were spawned. They also proposed an alternative method using the "z-axis" (elevation) to determine the sequence.
- Decoding: While the sender manually creates the message, the authors wrote a custom LUA script for the receiver that automatically scans the server's object list and decodes the RGB values into text.
Datasets and Viability
The paper does not use traditional training datasets. However, it includes a survey of Garry's Mod server statistics over a three-day period in May 2012 to establish the viability of the channel.
As shown in the data, there was a consistent environment of roughly 270–280 active servers with over 5,000 available slots, suggesting ample "room" to hide communication within the noise of active gameplay.
Key Insights and Discussion
- The "Nothing is too Weird" Rule: One of the strongest defenses of this channel is the culture of sandbox gaming. Because Garry’s Mod is designed for building strange contraptions, a player spawning dozens of colored barrels is rarely questioned by admins or other players.
- Observability vs. Proximity: Because the game client must know the state of every object on the server to render the world, a "receiver" does not need to be standing near the "sender" in the game. They simply need to be logged into the same server; their client will receive the object data automatically.
- Digital-Physical Hybrid: The authors argue this merges the two worlds. It is digital because it uses network protocols, but "physical" because the message is constructed by manipulating tangible (within the simulation) objects.
- Dead-Drops in Data: If a server allows props to persist after a player logs out, the message becomes a "storage-based" channel or a digital dead-drop.
Experimental Results
- Manual Transmission Speed: A user manually encoding a message ("Covert Channel Example") achieved a rate of 0.159 Bytes per second (1.275 bps).
- Pre-loaded Transmission Speed: By using the game’s "copy and paste" mechanics (pre-building a message offline and spawning it all at once), the bitrate jumped to 26.5 Bytes per second.
- Noise Mitigation: The authors found that engine artifacts (like "gibs" from destroyed objects fading out) could create temporary noise in the Alpha channel, but this was easily filtered by choosing specific "key" values.
Limitations and Future Work
- High Barrier to Entry: Both the sender and receiver must have the game installed, which requires a relatively high-end CPU and GPU. This makes it impractical for low-power or highly restricted environments.
- Network Visibility: Many corporate environments block the custom TCP ports used by gaming engines, which would prevent the game client from connecting to an external server.
- Manual Labor: Currently, there is no API to programmatically "inject" these messages; a human must still log in and interact with the game world.
- Future Directions: The authors suggest developing an API for automated manipulation and exploring map steganography, where data is hidden within the geometry of a custom-built game map itself, potentially allowing for gigabytes of hidden data.
Summary updated