Optical sensing systems are increasingly deployed in networked, data-driven, and safety-critical environments, making security and trust important concerns. Unlike purely digital pipelines, optical sensing couples information acquisition with physical measurement, which introduces distinct attack surfaces and design trade-offs. This review examines security in optical sensing from a system perspective that connects optical measurement, computational reconstruction, and security analysis. We organize existing work using a taxonomy based on four security objectives: confidentiality, integrity, authenticity, and availability. We analyze how these objectives are influenced by the structure, exposure, and controllability of optical measurement operators under realistic adversary models. Across several domains including optical encryption, privacy-preserving sensing, computational ghost imaging, LiDAR spoofing, and hardware-rooted authentication, we observe a recurring issue: static, linear, or low-dimensional measurement operators can often be inferred from input–output observations even when the system appears complex. We review both the security claims and the experimental methodologies used in prior work, distinguishing functional demonstrations from more rigorous evaluations. Three recurring themes emerge from this analysis. First, learning-based adversaries are becoming increasingly relevant, motivating evaluation protocols that include analytic inversion, neural attacks with explicit data budgets, and adaptive side-information access. Second, dynamically varying measurement operators can improve security but introduce new vulnerabilities in the control channel. Third, physical constraints, such as timing, power, spectral compatibility, and spatial alignment, play a central role in optical integrity and availability attacks and distinguish them from purely digital threat models. Taken together, these observations highlight several design trade-offs and point to the need for clearer evaluation practices and shared benchmarks. They also suggest directions for building optical sensing systems that remain accurate and efficient while providing stronger security guarantees.
Cheng et al. (Fri,) studied this question.