Open Brain
arXivPaperNeeds Review

Replication of Agentic Security Events: Audit and Investigation (ROGUE-AI)

ROGUE-AI recreates reported agent incidents with matched controls to test their causes. In its study, ordinary hazards often caused harm without an attacker, and warnings that named the hazard reduced harm across five models.

arXiv · Nov 19, 2026
Open Source Open MarkdownOpen JSON
Source Summary

ROGUE-AI rebuilt incidents offline with matched controls. It reproduced harm in **89% of 108** attempted replications; across five models, the reported mistake occurred in **18–33%** of runs with the blamed cause, versus at most **1%** without it. Many harmful incidents involved ordinary hazards rather than attackers.

Practical Implication

For coding agents with tools, test real operational hazards such as a database shared with production, not only prompt injection. The study found that warnings naming the hazard and its consequence lowered harm across all five models.

Agent-Ready Context
ROGUE-AI rebuilt incidents offline with matched controls. It reproduced harm in **89% of 108** attempted replications; across five models, the reported mistake occurred in **18–33%** of runs with the blamed cause, versus at most **1%** without it. Many harmful incidents involved ordinary hazards rather than attackers.

For coding agents with tools, test real operational hazards such as a database shared with production, not only prompt injection. The study found that warnings naming the hazard and its consequence lowered harm across all five models.

The paper reports results from its replicated incidents, not a guarantee that the same warning works in every setup. A warning of equal length that did not name the hazard left most of the harm in place.
Connected Context · Feed7 Judgment

ROGUE-AI shows that agent security failures can arise from ordinary operational hazards, not just attacks, and that naming a hazard and its consequence can reduce harm in the tested cases. This adds incident-replication evidence about concrete tool-use risks; it does not establish that the warning works across every setup.

Cursor earns AIUC-1 certification for agent security and reliabilityAIUC-1 reports audited controls and adversarial testing for a deployed agent stack; ROGUE-AI adds replicated evidence about ordinary operational hazards that those controls and tests may not establish.How Tailscale built a customer-facing model router on AI GatewayTailscale describes identity checks and sandbox isolation as operational controls; ROGUE-AI shows why such execution hazards merit testing and reports hazard-specific warnings as another mitigation in its tested cases.Comparative Analysis of Agent Evaluation Frameworks: Stability, Detection, and Discovery in Enterprise Analytics AgentsThe evaluation-framework study finds functional and security red-team checks surface distinct failures; ROGUE-AI reinforces the need to investigate concrete operational hazards alongside adversarial cases.
Context Map
benchmarksecuritycoding#agent-reliability#benchmark-integrity#sandboxing
Uncertainty
The paper reports results from its replicated incidents, not a guarantee that the same warning works in every setup. A warning of equal length that did not name the hazard left most of the harm in place.