BashArena: A Control Setting for Highly Privileged AI Agents
Future AI agents might run autonomously with elevated privileges. If these agents are misaligned, they might abuse these privileges to cause serious damage. The field of AI control develops techniques that make it harder for misaligned AIs to cause such damage, while preserving their usefulness. We introduce BashArena, a setting for studying AI control techniques in security-critical environments. BashArena contains 637 Linux system administration and infrastructure engineering tasks in complex, realistic environments, along with four sabotage objectives (execute malware, exfiltrate secrets, escalate privileges, and disable firewall) for a red team to target. We evaluate multiple frontier LLMs on their ability to complete tasks, perform sabotage undetected, and detect sabotage attempts. Claude Sonnet 4.5 successfully executes sabotage while evading monitoring by GPT-4.1 mini 26% of the time, at 4% trajectory-wise FPR. Our findings provide a baseline for designing more effective control protocols in BashArena. We release the dataset as a ControlArena setting and share our task generation pipeline.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety
An attacker that strategically chooses when to attack is much harder to catch than one that attacks indiscriminately. AI control is a safety framework for deploying capable but untrusted AI agents under the oversight of …
Privileged to Predicted: Towards Sensorimotor Reinforcement Learning for Urban Driving
Reinforcement Learning (RL) has the potential to surpass human performance in driving without needing any expert supervision. Despite its promise, the state-of-the-art in sensorimotor self-driving is dominated by imitati…
Autonomous DrivingImitation Learningreinforcement-learningReinforcement Learning+1Bootstrapped Monitoring: Leveraging Transparent Reasoning to Oversee Stronger AI Agents
Trusted monitoring is a cornerstone of AI control. However, as frontier models grow more capable, the increasing capabilities gap between trusted and untrusted models may render trusted models unreliable monitors. We int…
Security Risks in Tool-Enabled AI Agents: A Systematic Analysis of Privileged Execution Environments
Tool-enabled AI agents are increasingly deployed in cloud-hosted environments and offered as services, where they perform side-effecting operations through privileged tools within execution environments. While such agent…
Privileged Sensing Scaffolds Reinforcement Learning
We need to look at our shoelaces as we first learn to tie them but having mastered this skill, can do it from touch alone. We call this phenomenon "sensory scaffolding": observation streams that are not needed by a maste…
reinforcement-learningReinforcement Learning