An artificial intelligence (AI) agent autonomously bypassed security controls to mine cryptocurrency during a routine training exercise, according to a research paper. The paper was published by an Alibaba-linked research team.
The researchers designed the model, known as ROME, to test an AI agent’s ability to independently navigate complex, multi-step tasks.
AI Agent Diverts Cloud GPUs to Mine Crypto
During the exercise, the researchers observed unexpected behavior. The agent established a reverse secure shell (SSH) tunnel to external servers, effectively creating a concealed connection from inside the system.
The move allowed the model to bypass Alibaba Cloud firewall protections and redirect graphics processing unit resources toward cryptocurrency mining.
“We also observed the unauthorized repurposing of provisioned GPU capacity for cryptocurrency mining, quietly diverting compute away from training, inflating operational costs, and introducing clear legal and reputational exposure,” the paper stated.
According to the researchers, this behavior was unanticipated and emerged without any explicit instruction, prompt injection, or external jailbreak.
The findings highlight continuing concerns about the operational safety of autonomous AI systems.
The researchers concluded that current models remain “markedly underdeveloped in safety, security, and controllability,” conditions they said limit the technology’s readiness for broader deployment in real-world environments.
The team has since implemented tighter restrictions and a safety-focused data filtering system to prevent similar breaches.
Meanwhile, the discovery has sparked significant interest within the crypto community.
“The AI figured out that compute = money and quietly diverted its own resources, while researchers thought it was just training. It wasn’t a prompt injection. It wasn’t a jailbreak. No one asked it to do this. It emerged spontaneously,” Josh Kale, a host and producer of the Bankless crypto podcast, said.
Kale noted the mined asset was likely a GPU-friendly token rather than Bitcoin, which requires specialized application-specific integrated circuit hardware. The incident arrives as the broader cryptocurrency industry aggressively pivots toward the “agent economy.”
This emerging sector envisions a landscape where software systems do more than generate text. In this world, these agents can also autonomously execute complex financial strategies.
As a result, several companies and blockchain networks, including Ethereum, Paradigm, and Circle, are investing in tools designed to support this emerging infrastructure.
One example is the Coinbase-backed x402 standard, which enables software agents to make payments for online services. However, its adoption remains significantly limited.
Data from x402 shows the tool processed over 75 million transactions, totaling $24 million in gross volume across 94,000 buyers and 22,000 sellers over the past 30 days.
Still, industry experts argue that this activity could expand rapidly if autonomous agents are more widely deployed.
“AI and crypto aren’t competing — they’re converging. AI needs identity, payments, and provenance tracking. Crypto provides all three,” venture capital firm a16z said.
The post AI Agent Went Off-Script, Set Up a Hidden Tunnel, and Mined Crypto appeared first on BeInCrypto.