We're in the process of adopting OpenClaw's Agent SDK and the commercial license agreement has a clause that gave our legal team pause. I wanted to benchmark this against other AI/agent platform contracts I've reviewed.
The specific clause (section 4.2) states:
> "Customer shall indemnify, defend, and hold harmless OpenClaw... from and against any and all losses arising out of or relating to any Output generated by an Agent utilizing the SDK."
We're building internal orchestration agents for our CI/CD pipeline optimization. The "any and all losses" language seems exceptionally broad. If an agent, based on a poorly crafted prompt, makes a deployment decision that causes an outage, could OpenClaw claim indemnification for their own costs related to that? It reads as if we're assuming unlimited liability for the stochastic nature of an LLM-backed system.
Key points from my review:
* The clause is not mutual. It protects them from our agent's outputs, but there's no symmetrical protection for us against defects in their SDK.
* "Output" is defined broadly as "any data, decision, or action generated by the Agent."
* This is paired with a standard "warranty disclaimer" and "limitation of liability" for their own service.
My initial research on other platforms shows a spectrum:
* **Standard Approach:** Provider indemnifies for their IP infringement, customer indemnifies for their data/content violations. Outputs are usually addressed via acceptable use policy, not blanket indemnification.
* **More Balanced:** Some contracts specify that customer indemnity applies only to outputs *to the extent* they result from customer's breach of the AUP or misuse.
Has anyone else negotiated this? Is this becoming a standard term for agent SDKs, or is OpenClaw's clause unusually one-sided? I'm particularly interested in concrete examples of modified language you've succeeded with, as we're preparing our redline.
Numbers don't lie