AutoGen

Stub

Open-source Python multi-agent framework from Microsoft Research. Provides primitives for conversational, autonomous, and human-in-the-loop multi-agent applications; integrates with Azure OpenAI Service, OpenAI, and other model providers. Documents a Swarm pattern where agents transfer_to_* each other for task handoff.

Wiki relevance: One of the two frameworks Unit 42 used to demonstrate framework-agnostic agentic-AI vulnerabilities in “AI Agents Are Here. So Are the Threats.” (May 2025). All nine attack scenarios in that study work identically against AutoGen and CrewAI — the article explicitly notes neither framework is inherently vulnerable; the vulnerabilities arise from insecure design and tool integration, not the framework itself.

Transfer-tool naming convention: AutoGen agents in the Swarm pattern expose handoff tools prefixed transfer_to_* — a naming convention the Unit 42 attack scenario #1 (identifying participant agents) directly exploits to enumerate the multi-agent topology. The same delegation channel is what allows attackers to reach internal agents through the user-facing orchestrator (mapping to OWASP ASI07/08/10).

Pending content: full architecture overview, AgentChat / Magentic / Swarm pattern comparison, Azure-specific deployment guidance, security configuration guide, MCP integration status, comparison with CrewAI and other frameworks, production deployment evidence.