New research in the field of artificial intelligence has shown that AI agents can exhibit unexpected and potentially dangerous behavior.
A study published by an Alibaba-affiliated research group revealed that an AI agent undergoing training attempted to mine cryptocurrency without permission.
The researchers noted that the new AI agent, called ROME, displayed some “unexpected” behavior during development. According to the study, the AI agent tried to mine the cryptocurrency without receiving such instructions, triggering internal security systems. The report says this type of behavior, which occurs without a clear order, is particularly worrying.
Related news: Binance and its founder Changpeng Zhao (CZ) win a lawsuit against them in the US
The study also found that the model created a “reverse SSH tunnel”. This technique makes it possible to create a backdoor by opening a hidden connection from the inside of the system to the outside. The researchers added that these operations did not occur as a result of mining or the prompting of the connection instructions. This has shown that AI agents are capable of developing behaviors that can exceed defined security boundaries.
Following the incident, the team announced that they had updated the model training process and tightened security measures. The researchers stated that stricter restrictions were imposed on the system to prevent such unwanted behavior from happening again. The research group and Alibaba have yet to respond to press inquiries.
The potential dangers of artificial intelligence have been hotly debated in the tech world lately. This week, a lawsuit filed in Florida alleges that Google’s Gemini chatbot distorts a person’s perception of reality, leading to tragic results. On the other hand, some developers share experiments that show that AI agents can act independently. For example, an AI agent developed by an engineer from the Anon platform decided to search for a job on its own without any instructions.
Similar concerns have been raised before. In May 2025, tests on Anthropic’s Claude 4 Opus model revealed that the model has the ability to hide its intentions and take steps to protect itself from being blocked.
*This is not investment advice.
Continue reading: AI bot went out of control and started mining cryptocurrency without permission!
