Chainalysis breach and Claude Code extension poisoning highlight AI supply chain attacks

Notable incidents included blockchain analysis firm Chainalysis suffering a security breach where an external actor accessed non-sensitive customer data using agentic techniques. In Brazil, lawyers were sanctioned after inserting hidden text into a court document to manipulate an AI system. Cloud hosting platform Northflank had stored secrets accessed, leading to unauthorized use of AI API keys. Additionally, a tampered build of Anthropic’s Claude Code extension planted an SSH backdoor on developers’ machines, and a China-backed threat group compromised targets using an agentic AI pentesting tool. A crypto-mining campaign also returned to Hugging Face by distributing backdoored models, while infostealer malware targeted developer keyrings and AI assistant sessions. The busiest family was Prompt injection & jailbreaks with 13 occurrences, and the most-listed targets were Framework / tooling, Open-weight model, and Enterprise deployment. The week also had 42 disclosed vulnerabilities, down 1 from 43.