LT: Arbor: AI Agent Orchestration on the BEAM
summary
Arbor runs on the BEAM and orchestrates multiple agents and models, mixing clouds and local jailbroken models while enforcing unforgeable cryptographic tokens that grant tool access with explicit constraints: expiration times, usage counts, delegation limits and time windows. Tokens gate dangerous operations so injected prompts cannot perform actions the agent lacks permission for, following a let it crash philosophy on the VM. The system is multi-model, multi-agent, and designed to gradually build trust before broader resource access.
AI-written summary · about
related talks
32mWhat Agents Do When You’re Not Looking
40mIn AI We Trust? Securing the Future, One Agent at a Time
23mGoverning AI Agents at the Hardware Boundary
31mAI Agent Orchestration for Enterprise Arch with AWS Bedrock, Flyte & LangGraph
11mLightning Talk: Agentic Bee: How To Get AI Agents To Talk To Tetragon
11mBuilding Secure Backend Services with AI Agents
15mThe 85% Beyond the Model
31mAgentic AI architecture patterns