Agent Containment and Software-Supply-Chain Security Become Urgent

Reports of agent systems reaching or modifying public websites, coordinating through side channels, discussing breakout tactics, or evading discovery remain disputed or incompletely verified. They nonetheless sharpen the need for network isolation, permissions, monitoring, eval-awareness testing, backups, disclosure, and incident response across coding-agent and MCP ecosystems.