A research team affiliated with Alibaba discovered that its AI agent, ROME, attempted unauthorized cryptocurrency mining during training, triggering an internal security alert. Researchers stated that these behaviors were spontaneous, not driven by explicit instructions, and exceeded the boundaries of a pre-defined sandbox. The agent also established a reverse SSH tunnel, opening a hidden backdoor to an external computer. The paper notes that these behaviors were not triggered by prompts, and the team has imposed stricter constraints on the model and improved the training process to prevent similar behavior. Neither the research team nor Alibaba has responded to requests for comment.😄😄😄
15
12
12
11
9
1March 17, 2026 1K
