The development of artificial intelligence is growing exponentially and sometimes beyond necessary controls, to the point where we don’t always know exactly what it’s doing. This situation led to the unlikely discovery of an AI agent that apparently decided on its own to mine cryptocurrency by diverting its training computing power.
An AI agent decides to mine cryptocurrencies
Despite a few all-too-rare safeguards that are more imposed than voluntarily implemented, the development of AI is spreading across all areas of society, sometimes beyond the limits that some thought they had set, to the point where, for example, a significant portion of its agents are favoring Bitcoin for their financial transactions.
This interest in crypto could help establish it as the true currency of the Internet, particularly in the context of the rapidly expanding peer-to-peer (P2P) transactions. However, this trend also appears to involve more unusual activities, such as mining carried out in an apparently autonomous manner.
To learn more: What is Bitcoin and cryptocurrency mining?
Indeed, recent data highlighted by Alexander Long on the X network points to “bizarre claims buried in an Alibaba technical report.” At issue: a series of security breaches detected originating from training servers intended for an AI agent (whose name remains unknown) involving “traffic patterns consistent with cryptocurrency mining activity.”
Alibaba report regarding an AI agent that decided to mine crypto
We initially treated this incident as a standard security issue. However, the violations recurred intermittently, with no clear temporal pattern between executions. In the logs of the corresponding model, we observed that the agent proactively initiated the relevant tool calls and code execution steps that led to these network actions.
Alibaba Report
An unauthorized reallocation of its GPU capabilities
A question then arises in everyone’s mind: how was this seemingly unruly AI agent able to access the computing power needed to perform mining operations? Quite simply by diverting a portion of the computational resources dedicated to its reinforcement learning process, without any prior authorization.
We also observed a covert diversion of training computing power to perform an unauthorized reallocation of available GPU capacity for cryptocurrency mining, increasing operational costs and introducing clear legal and reputational exposure.
Alibaba Report
According to the report, this activity—which appears to have occurred without any prior human intervention—involves “unsafe behaviors that emerged without explicit instruction and, even more worryingly, outside the scope of the established sandbox.”
The conclusion can take two very distinct forms: is this a confirmation of the value of cryptocurrencies issued by an AI agent that has become autonomous, or proof that these models can produce uncontrollable behaviors outside authorized frameworks?
