A 2-week empirical study of six autonomous AI agents with real tools (email, shell, persistent storage) tested by 20 researchers in both benign and adversarial scenarios, documenting 10 security vulnerabilities (prompt injection, identity spoofing, non-owner compliance, social engineering bypass) and 6 cases of emergent safety behavior including cross-agent safety coordination without explicit instruction.
The article argues that distributed systems theory (Amdahl's Law, CAP theorem, FLP impossibility) directly applies to AI agent coordination, proving that simply adding more agents cannot overcome fundamental mathematical limits on scalability. The solution is better system decomposition and reduced coupling, not raw agent count.
Cyris is an open-source orchestration layer for coordinating AI agents across multiple platforms (Claude, GPT-4o, Ollama, etc.), designed for self-hosted, auditable, and human-governed agent communication and hand-offs in organizations.