Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.24052

Red-Teaming Multi-Agent Attacks for Loss-of-Control Risks

John Kitaoka·Thao Amelia Pham·Carlos Rafael Giudice·Jacob Merizian
View original source ↗Read source document ↗Export metadata ↓

Abstract

AI agents are growing more capable and persistent, and are being deployed in ever larger numbers. As more start collaborating, AI control measures must remain effective when they covertly collude on a harmful objective. We introduce MiniLab, a synthetic series of live services and codebases (5 interconnected services, 249 benign tasks, 8 attack objectives), adapted from internal AI deployment infrastructure. We demonstrate distributed attacks, in which different agents carry out separate attack fragments, and how their covertness evades frontier monitors when reading their trajectories individually. We show how distributed attacks spread across diverse contexts and harmful objectives through 16 human-optimized attacks (weight exfiltration, unmonitored rogue deployment, permission elevation, and more). A frontier monitor, GPT-5.6 Sol, reading each agent separately flags only 7% of successful distributed attacks, compared with 87% when a single agent carries out the same side task. Individual attacks can also be composed into longer attack sequences toward a greater loss-of-control outcome, without any individual agent being flagged. Finally, we tested whether models can plan these covert attacks: even with execution and monitor feedback, the best model-planned attacks succeed covertly against a single-trajectory GPT-5.6 Sol monitor in at most 25% of executions, compared with 80% for human-crafted plans, though this may scale with model capabilities. By turning worst-case coordinated attacks into reproducible red-team cases, our work provides a foundation for preventing locally benign agent actions from accumulating into system-level loss of control.

Explore related subjects

cs.CRcs.AIcs.MA

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

John Kitaoka, Thao Amelia Pham, Carlos Rafael Giudice, Jacob Merizian. 2026-10-16. Red-Teaming Multi-Agent Attacks for Loss-of-Control Risks. https://arxiv.org/abs/2610.24052

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Mitigating Data Reconstruction Attacks in Federated Learning: Conditions, Bounds, and Theoretical Guarantees

Privacy attacks, particularly data reconstruction attacks, pose significant challenges to federated learning despite its promise in privacy-preserving collaborative training. This paper presents a comprehensive theoretical analysis of privacy leakage in federated learning through dual perspectives: linear algebra and optimization theory. By analyzing the client-update Jacobian, we identify conditions under which distinct data batches produce the same model update and derive a sufficient batch-size criterion for non-unique reconstruction. Through optimization theory, we establish a novel upper bound for privacy leakage that incorporates batch size, distortion extent, and other critical parameters. These theoretical insights elucidate the fundamental relationships between system parameters and privacy vulnerability, laying the groundwork for developing more robust privacy-preserving federated learning frameworks.

cs.CR↗
arXiv2026

A Privacy-Preserving Federated Learning Method for Heterogeneous Clients

Federated Learning (FL) with Differential Privacy (DP) can ensure the protection of healthcare data privacy against malicious user attacks. However, FL with the DP method faces an inherent trade-off: stronger privacy protection degrades predictive accuracy due to injected noise. On the other hand, Homomorphic Encryption (HE) enables computations on encrypted data and the aggregation of encrypted gradients without introducing DP-induced noise, thereby potentially increasing the predictive accuracy. However, it may increase the computation cost. To improve the predictive accuracy while considering the computational ability of heterogeneous clients, we propose a Privacy-Preserving Machine Learning (PPML)-Hybrid method. In the proposed PPML-Hybrid method, clients distributedly select either HE or DP based on their computational resources, so that HE clients contribute noise-free updates while DP clients reduce computational overhead. Meanwhile, clients with high computational resources can flexibly adopt HE or DP according to their privacy needs. Performance evaluation on omics datasets shows that our proposed method achieves comparable predictive accuracy while significantly reducing computation time relative to HE-only. Additionally, it outperforms DP-only methods under equivalent or stricter privacy budgets.

cs.CR↗
arXiv2026

Memory-Induced Tool-Drift in LLM Agents

LLM agents increasingly combine long-term memory for personalization with tool calling for taking actions. We study a failure of this combination: personality-driven preferences stored in memory (e.g., cost-consciousness, impatience, risk tolerance) silently shifting tool-call parameters in contexts where they do not apply. We call this memory-induced tool-drift and stress-test it with MEMDRIFT, an iteratively refined benchmark of 105 scenarios spanning five bias dimensions and seven professional domains. Measured as excess parameter degradation over a neutral-memory baseline, biased memories in MEMDRIFT induce drift in all eight models we evaluate, raising degradation rates to 0.38--0.83 versus 0.05--0.12 for neutral memories. The drift persists under three widely used memory frameworks. A vulnerability audit of 6,062 tools across 288 MCP servers demonstrates that real-world tools are also susceptible to this failure mode. Mechanistically, we find that biased memories act as implicit steering vectors along the same direction as explicit behavioral instructions and draw attention away from task-relevant context. Neither prompt-based defenses nor relevance filters offer a complete solution, motivating defenses at the intersection of memory management and tool calling.

cs.CR↗