On August 30, 2026, Web3 project co-founder Numa Lunah narrowly avoided losing control of his digital assets after an AI-assisted setup process delivered a hidden infostealer to his work laptop.
The incident began when Lunah asked Claude, an AI assistant, for a download link to a transcription application. The link led to a phishing clone of the official site. After installation, malware designed to steal passwords, exchange credentials, and private keys from hot wallets was silently deployed.
Lunah detected the compromise, isolated the device, and performed a complete operating system reinstall. However, the threat was not eliminated. While restoring files from backup, he found that hackers had altered a SKILL.md document used as a personal AI style guide. When the modified configuration file was added to a clean machine, it automatically connected to an attacker-controlled server, downloaded the infostealer again, and resumed credential collection.
NEAR Protocol co-founder Illia Polosukhin warned that securing autonomous AI agent infrastructure is becoming critical and highlighted a growing number of campaigns using context poisoning tactics. The case underscores that AI skill configuration files in .md or .json formats must be treated like executable code and thoroughly inspected before use.
