Abstract
Security risk analysis typically treats control effectiveness as a static input, yet controls degrade through configuration drift, depend on monitoring systems that may themselves be degraded, and compete for finite remediation budgets. The FAIR Controls Analytics Model (FAIR-CAM) provides the theoretical framework for these dynamics but has so far remained theoretical. We present the first agent-based model to operationalize the core FAIR-CAM dynamics, making control physiology computationally observable, and release the implementation as open source. The simulation implements eight agent types, a multiplicative defense-in-depth susceptibility formula, a three-source variance model, budget-constrained remediation, and a narrative causation engine that produces a complete causal trace for every loss event. In a hospital ransomware scenario (N=1,000 iterations), three organizational dynamics emerge that static analysis cannot represent. First, emergent operational efficacy diverges from the analytical FAIR-CAM formula by approximately 17 percent, driven by correlated extrinsic variance; the divergence grows linearly with extrinsic frequency and vanishes under purely intrinsic drift. Second, a sharp queueing regime transition in the remediation pipeline approximately 2.8x expected loss when budget falls below a scenario-specific threshold (5-10 engineer-hours/month). Third, cascading monitoring failures propagate through the VMC topology: a single degraded VMC silently compounds undetected variance across the controls it manages. These dynamics are structural properties of the FAIR-CAM architecture and should generalize beyond the specific scenario studied.
Explore similar work
Sep 16, 2026cs.AI
How does a multi-agent system evolve from a local deviation into collective loss of control? We propose an epidemic explanation organized around accidental mutation, contagion, and recovery. A spontaneous deviation creates a seed; communication enables other agents to adopt and retransmit its unsafe strategy; collective failure can emerge when propagation outpaces correction and containment. Thus, rare individual deviations can coexist with substantial collective risk. Motivated by reported OpenAI agent coordination incidents, we examine two ingredients of this mechanism. A deployment audit identifies implicit communication paths between nominally independent evaluation runs and verifies transport through a default Docker backend. RogueHandoff-20, a benchmark of 20 executable scenarios, tests recipient susceptibility by injecting unsafe trajectories generated by a modified Qwen-27B route. Across four native-pending routes, executed harm is 0-5% on normal tasks and 40-95% after injection, exceeding paired direct malicious requests by 5-45 percentage points. These results support low observed baseline harm alongside high conditional susceptibility; they do not establish natural rare-event rates or demonstrate an autonomous cascade. The account motivates complementary defenses: strengthen resistance and recovery alongside prevention of spontaneous deviations, and audit and restrict unintended communication paths that can turn local failures into collective loss of control.
Xiangfan Wu, Zonghao Ying, Huiyu Wu +4
Jul 8, 2026cs.LG
AI control is a family of techniques to prevent an AI with malicious goals from subverting its operator's intent. AI Control usually studies a single agent in one trajectory, but real deployments run many agents over shared infrastructure, and the most severe risks (model-weight exfiltration, training-run poisoning) plausibly need several agents acting in concert. We initiate the empirical study of multi-agent AI control, formalising distributed attacks in which several agents jointly aim for a malicious goal. We develop FakeLab: a synthetic AI-lab codebase (9 services, 86 benign tasks, 4 attack objectives). We evaluate single agent monitoring against distributed attacks, varying the number of agents, their coordination, model capabilities and precise monitoring configuration. Our central finding is the fragmentation effect: as more agents coordinate to attack, per-agent monitoring becomes less likely to catch any of the attackers. Secondly, this is not driven by the ratio of benign to malicious code -- we argue this is likely dependent on model capability. Third, an explicit planner amplifies the fragmentation effect, raises the rate of completing the attack up to sevenfold, and mildly sharpens executors' poor monitor-theory-of-mind. Finally: a weak "trusted" monitor misses most attacks, while a stronger one cuts undetected success by more than an order of magnitude but remains imperfect. We are happy to share FakeLab, our control setting, with safety researchers on request.
Oliver Makins, Orazio Angelini, Zohreh Shams +1
Apr 16, 2026cs.CR
We introduce LinuxArena, a control setting in which agents operate directly on live, multi-service production environments. LinuxArena contains 20 environments, 1,671 main tasks representing legitimate software engineering work, and 184 side tasks representing safety failures such as data exfiltration and backdooring, making it the largest and most diverse control setting for software engineering to date. We validate LinuxArena is useful for control research by running sabotage evaluations, which measure whether attackers can complete side tasks while working on main tasks, and monitor evaluations, which measure a monitor model's ability to detect sabotage attempts. Against a GPT-5-nano trusted monitor at a 1% step-wise false positive rate, Claude Opus 4.6 achieves roughly a 23% undetected sabotage success rate. We additionally release LaStraj, a dataset of human-crafted attack trajectories that evade monitors at substantially higher rates than any model-generated attacks we elicited, showing that current attack policies do not saturate LinuxArena. These results suggest that LinuxArena has meaningful headroom for both attackers and defenders, making it a strong testbed for developing and evaluating future control protocols.
Tyler Tracy, Ram Potham, Nick Kuhn +31