Three big vendors shipped the same idea on the same day: AI agents need fences. Here’s the quick rundown.
Akamai: a place for agent code to run
Akamai launched Agent Sandbox (beta), an open source part of the Spin framework that uses WebAssembly isolation to cap compute, memory, outbound connections and system capabilities for the code agents write and run. It also launched Akamai Inference for serving open-weight models.
IBM: every agent gets an ID
At TechXchange, IBM announced Agent Identity in HashiCorp Vault, giving each agent a governed identity with least-privilege secrets and real-time revocation. watsonx Orchestrate is meant to discover, contain and evaluate agents across Amazon Bedrock, Microsoft Foundry and Google Gemini.
Salesforce: catch the runaway
Salesforce added Agent Evals and API Evals (beta in November), cost caps, and an Agent Kill Switch that’s generally available today.
My take
Sandbox, identity, kill switch. That’s the same three-layer setup I want on my own bots: run them somewhere they can’t hurt the box, give each one its own scoped keys, and keep a big red button handy. Glad the enterprise world is catching up.
God Bless, and May the Source Be with You!
