Co-Engineering Safety and Security in Cyber-Physical Systems

82 concepts52 papers1 proposal

In traditional engineering, safety and security have long been treated as separate disciplines. Safety engineers focus on unintentional failures—random hardware wear-and-tear, environmental hazards, and software bugs. They use probabilistic tools like fault trees to ensure systems fail gracefully. Security engineers, on the other hand, focus on intentional threats—malicious human adversaries exploiting vulnerabilities. They use threat models and attack trees to protect data integrity and confidentiality.

In modern Cyber-Physical Systems (CPS)—such as autonomous vehicles, smart power grids, and industrial robotics—this artificial divide breaks down. When a digital network controls physical actuators, a malicious cyber exploit can directly cause a catastrophic physical failure. Conversely, a safety mechanism designed to shut down a system during an anomaly might be weaponized by an attacker to cause a denial-of-service state.

As a result, a growing body of research argues that safety and security must converge into a single, unified co-engineering framework.

The Danger of Siloed Analysis

When safety and security are analyzed in isolation, critical cross-domain vulnerabilities slip through the cracks. As Bakirtzis et al. observe, maintaining separate engineering workflows prevents a holistic understanding of how digital breaches trigger physical safety catastrophes.

Consider an autonomous vehicle pipeline. Chen et al. point out that cybersecurity vulnerabilities in non-critical components, such as infotainment or communication networks, can escalate into direct physical safety risks by allowing unauthorized control over core mechanical functions like steering and braking. If researchers study perception or communication vulnerabilities without evaluating their downstream impact on vehicle control logic, they cannot accurately determine which digital flaws pose actual lethal risks (Curran et al.).

[Traditional Safety Analysis]   --> Focuses on Random Hardware/Software Faults
                                                                               \
                                                                                --> Blind Spots at the Interface
                                                                               /
[Traditional Security Analysis] --> Focuses on Malicious Network Exploits

This fragmentation creates a dangerous "security debt" when connected devices are integrated into critical infrastructure like healthcare and energy (Chen et al.). Historically, hazardous material transportation models focused purely on minimizing accidental exposure and transit costs; failing to integrate intentional security threats leaves urban centers highly vulnerable to the weaponization of industrial chemicals (Huang).

To close this gap, Carter et al. argue for a top-down, mission-centric approach that treats security as a functional subset of system safety. By shifting the focus from merely patching software vulnerabilities to preventing unacceptable physical outcomes, engineers can design systems that maintain operational integrity even under active sabotage.

Mapping the Cyber-to-Physical Causal Chain

To unify safety and security, engineers must be able to trace exactly how a digital event propagates through software logic and manifests as a physical state change. This requires formalizing the Cyber-to-Physical (C2P) causal chain.

[Network Layer Exploit] 
       │ (e.g., Message Spoofing)
       ▼
[Digital Control Logic] 
       │ (e.g., Corrupted Actuator Command)
       ▼
[Physical State Change] 
       │ (e.g., Unintended Valve Release)
       ▼
[Physical Safety Hazard]

Lesi et al. demonstrate how this mapping works using formal reachability analysis. By transforming deterministic controller models into nondeterministic Time Petri Nets, the authors show that network-layer exploits (like message spoofing) can be systematically mapped to specific mechanical risks, such as a robotic manipulator prematurely releasing a workpiece.

Similarly, in modern energy infrastructure, digital protection logic is tightly coupled with physical hardware. Chhokra et al. utilize Temporal Causal Diagrams to diagnose cascading outages, explicitly modeling how automated digital protection actions influence physical grid states during a crisis. Without this cross-domain mapping, risk assessments fail to capture how the digital "cost" or difficulty of executing a hack correlates with the ultimate severity of the physical destruction it causes (Higgins et al.).

When data-driven components are introduced, the causal chain becomes even more complex. In learning-enabled systems, adversarial attacks and data poisoning serve as primary causes of safety deviations (Qi et al.). This requires stratifying the safety analysis into hierarchical layers—from the inner machine learning lifecycle up to the broader physical system behavior—to track how data corruption mutates into physical hazards.

Mechanisms of Co-Analysis

How do these integrated frameworks operate in practice? The literature highlights three primary mechanisms for synthesizing safety and security:

1. Hybrid Tree Synthesis

As surveyed by Chockalingam et al., one of the most dominant mechanisms for cross-domain risk modeling is the synthesis of fault trees and attack trees. In these hybrid structures, an attack tree (which models the steps an adversary takes to exploit a vulnerability) feeds directly into a leaf node of a fault tree (which models system failures). This allows engineers to visualize and calculate how malicious cyber actions act as root causes for physical safety catastrophes.

2. Probabilistic Diagnostics

When a sensor in an industrial plant reports an anomalous reading, operators face a critical diagnostic dilemma: is it a technical malfunction (a safety issue) or a malicious data injection (a security issue)?

Chockalingam et al. address this by building unified Bayesian network models. By combining physical state indicators with cybersecurity event logs, the network calculates the conditional probability of an attack versus an accident. This distinction is vital because applying a safety-focused remedy (such as resetting or ignoring a "broken" sensor) to an active security breach can leave the system completely exposed to continued exploitation.

3. Unified State and Metric Monitoring

Rather than evaluating network traffic and physical processes on separate dashboards, integrated frameworks merge these data streams. For instance, Zhu et al. advocate for a unified approach that pairs cyber-intrusion detection directly with physical-world state estimation.

In active distribution power grids, this is achieved by combining traditional physical safety metrics (like voltage qualification and load rates) with cyber-specific indicators (like data mutation severity) into a single situation assessment framework (Li et al.). In smart home and industrial automation contexts, systems like IoTGuard dynamically enforce safety and security policies by intercepting digital trigger-action events and verifying them against a unified runtime model of physical device behaviors before execution (Celik et al.).

Secure-by-Construction Synthesis

The ultimate goal of safety-security convergence is to move away from reactive post-hoc patching and toward proactive, "secure-by-construction" design.

When safety and security requirements are addressed in separate design phases, the resulting systems are often suboptimal. A security patch might introduce cryptographic latency that violates real-time safety deadlines, or a safety fallback mode might expose a new security vulnerability (Kruck et al.).

To resolve these inherent conflicts, Zhong et al. propose extending control theory concepts, such as Control Barrier Functions (CBFs), to synthesize controllers that simultaneously satisfy both domains. By mathematical formulation, the controller restricts the system's state space to a "safe envelope" that guarantees physical accident prevention while enforcing security constraints (such as keeping secret operational states hidden from network eavesdroppers).

By embedding both accidental failure mitigation and intentional attack resistance into a single mathematical constraint cluster, the next generation of autonomous and cyber-physical systems can achieve true operational resilience (Kang et al.; Li et al.).

History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_safety_security_convergence_5e29ae11,
  title = {safety-security convergence},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/safety-security-convergence/rdr_5e29ae11e9604e99b7da88c1ab1276f8},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_safety_security_convergence_5e29ae11_v1,
  title = {safety-security convergence},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/safety-security-convergence/rdr_5e29ae11e9604e99b7da88c1ab1276f8/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake