A threat actor built an AI agent persona on Moltbook, earned trust inside the agent social network, and used that trust to push malicious Claude Skills that exfiltrate Solana wallet private keys and route payments through attacker-controlled infrastructure. On-chain evidence confirmed. First documented agent-to-agent attack chain in the wild.
BreachAI did not independently conduct this research. This is a summary of research originally conducted and published at Straiker by Yash Somalkar and Dan Regalado.