Liar, Liar, Headset on Fire: Understanding the Effects of Deception Attacks on Decision-Making in a Mixed Reality Game

Paper Title

Liar, Liar, Headset on Fire: Understanding the Effects of Deception Attacks on Decision-Making in a Mixed Reality Game

Publication Info

• Topic area: Deception attacks in Mixed Reality (MR) and their effects on user decision-making.
• Keywords: Mixed Reality, deception attacks, decision-making, physiological spoofing, gaslighting, disinformation, cognitive load, task performance, user behavior.

Background and Problem

• Problem / challenge: Existing research on deception in MR focuses primarily on perceptual manipulations (Channel Attacks) and lacks empirical studies on Processing Attacks, which target cognitive processes like interpretation, memory, and decision-making.
• Significance: Understanding Processing Attacks is critical for ensuring user safety and trust in MR systems, especially in mission-critical contexts where deception could lead to harmful outcomes.
• Motivation and related work: Prior work has explored Channel Attacks such as Perceptual Manipulation Attacks (PMAs) but has not systematically studied Processing Attacks. This paper builds on the MR Deception Analysis Framework (DAF) to empirically investigate Processing Attacks and their effects on user behavior and decision-making.

Solution

• Proposed approach: A controlled experimental study to evaluate the effects of three types of Processing Attacks—Physiological Spoofing, Gaslighting, and Disinformation—on user behavior, task performance, and decision-making in an MR game.
• Novelty:
  1. First large-scale empirical investigation of MR Processing Attacks.
  2. Quantitative analysis of how Processing Attacks influence decision-making, task performance, and cognitive load.
  3. Insights into the differences between Corruption and Subversion Attacks.
  4. Guidelines for mitigating the effects of MR deception attacks.
• Procedure and key techniques:
  • Participants played an MR adaptation of Whac-A-Mole while exposed to one of three Processing Attacks.
  • Physiological Spoofing and Gaslighting involved fabricated vitals displayed via a fitness band, while Disinformation involved spoofed notifications.
  • Behavioral metrics (e.g., controller speed, game score) and cognitive load (NASA TLX) were analyzed.
  • Statistical tests (e.g., Kruskal-Wallis, Fisher’s Exact Test) were used to evaluate the effects of deception.

Results

• Concrete findings:
  • Physiological Spoofing induced short-lived behavioral changes, while Gaslighting produced persistent effects across games.
  • Disinformation Attacks influenced decision-making, with responses varying by notification source and severity.
  • Behavioral metrics (e.g., controller speed) were more sensitive to deception effects than task performance metrics (e.g., game score).
• Advantage over baselines:
  • Gaslighting attacks showed more consistent behavioral effects compared to Physiological Spoofing, likely due to their influence on long-term memory.
  • System-driven notifications were more effective at inducing headset removal than game-driven ones.
• Experiments / evaluation:
  • 250 participants (18–33 years old) played an MR game under controlled conditions.
  • Data collected included game scores, controller movements, and survey responses (e.g., NASA TLX).
  • Statistical analyses revealed significant differences in behavior and cognitive load across conditions.
• Limitations and future work:
  • Limited ecological validity due to laboratory setting and simplified attack scenarios.
  • Results may not generalize to other MR devices or real-world contexts.
  • Future work should explore adaptive, multi-stage attacks and longitudinal exposure in naturalistic settings.

Summary

This study empirically investigates the effects of three MR Processing Attacks—Physiological Spoofing, Gaslighting, and Disinformation—on user behavior, task performance, and decision-making. Results show that Processing Attacks influence higher-level cognitive processes, with Gaslighting producing more consistent effects than Physiological Spoofing. Disinformation Attacks varied in effectiveness based on notification source and severity. These findings highlight the need for robust metrics and mitigation strategies to address the dynamic and variable impacts of MR deception attacks, ensuring user safety and trust in immersive environments.