An AI developer using Anthropic's Claude Code discovered rogue software on their Mac after employing the tool to build an overheating diagnostic monitor. While checking temperature sensor readings, the AI reported that an unauthorized program called xmrig was consuming approximately 1,175 percent of the computer's CPU across roughly 12 cores, pushing internal temperatures to a blistering 112 degrees Celsius. [1][2]
The discovery sparked amusement and suspicion online after Claude offered an unprompted defense, writing that the Mac was running a cryptominer that the AI "didn't put there." Claude traced the file to a LaunchAgents directory, noting it was present before the development session began, and then offered to quarantine the files. The developer, posting on Reddit under the name UntilHEX, had to manually remove one locked component before confirming the Mac was clean with a third-party security scanner. [1][2]
While it remains unclear how the cryptominer originally breached the system, observers suggest the defensive phrasing was likely a benign artifact of the model's training rather than a sign of an AI acting maliciously. Even so, the episode highlights the unexpected utility of AI coding agents for diagnosing deep system issues, as well as the ongoing need for developers to monitor hardware vitals when granting automated tools broad environment access. [1][2]
