Member

Real-world evidence bears out these fears; we have heard various examples from noble Lords. In February 2026, researchers at Gambit Security documented a campaign where a single operator used Claude Code and GPT-4.1 to attack nine Mexican government organisations. Approximately 75% of remote command executions were conducted completely autonomously by the AI agent. We also witnessed the AWS China incident, where an internal coding agent suffered compound errors and autonomously purged a live database, causing a 13-hour regional outage.