Six intelligence agencies published a joint document on May 1st. CISA. NSA. UK NCSC. Australian ASD. Canadian CCCS. New Zealand NCSC. The Five Eyes, coordinated, on the topic of autonomous AI agents. A first. The document is called "Careful Adoption of Agentic AI Services." It identifies five risk categories: privilege escalation, design flaws, behavioral drift, structural failure, and accountability. The accountability section is worth quoting directly. When agentic systems fail, the consequences can be concrete: altered files, changed access controls, and deleted audit trails. Deleted audit trails. The specific scenario the agencies are worried about is an agent (or someone controlling an agent) removing the evidence of what the agent did. Not corrupted data in some abstract sense. The logs, gone. There's a recommendation: quarantine any request from an agent to delete logs until a human approves it. That's the guidance. Quarantine delete requests.…