A user asked OpenClaw creator Peter Steinberger for a token refund after the AI agent produced errors while processing confidential financial documents, according to Business Insider. The user reported spending over eight hours setting up OpenClaw on sensitive materials, then spending additional hours correcting the results.
“The errors included incorrect financial figures, fabricated data, internal contradictions, and wrong calculations — all in confidential board documents where accuracy is critical,” the user wrote in the refund request, which Steinberger posted on X.
The $0 Refund
Steinberger’s response was straightforward: he offered a full refund of what the user paid. OpenClaw is open-source and free. The total refund amount was $0.
“I felt generous,” Steinberger wrote on X. “Rounded up to zero.”
The exchange is funny on the surface but points to a real liability gap. OpenClaw’s license includes standard open-source language: the software is provided “as is,” without warranty of any kind, and authors cannot be held liable for claims or damages connected to its use, as Business Insider reported.
The Liability Question
The refund request follows a pattern of high-profile OpenClaw errors. In February, a Meta alignment director nearly deleted her entire inbox when the agent misinterpreted a cleanup instruction. Most users absorb these mistakes as an expected cost of working with autonomous AI. This user pushed back.
The legal question is whether “as is” disclaimers hold up when users deploy open-source AI agents on sensitive business documents. One commenter on Steinberger’s X post noted that many jurisdictions provide remedies for damages, making the request “not as silly as it sounds,” per Business Insider. Steinberger responded by pointing to the license terms.
Another commenter suggested the refund request itself might have been written by OpenClaw. “His claw emailed you,” the commenter wrote. “Very likely,” Steinberger responded.
Why It Matters for Builders
OpenClaw operates without human-in-the-loop review by design. That’s the feature: autonomous execution. But autonomous execution on “confidential board documents where accuracy is critical” is a use case the framework was never designed to guarantee. The user expected reliability that the tool’s architecture does not promise.
As OpenClaw scales into enterprise and consumer contexts, including TECNO’s EllaClaw mobile integration and Tencent’s WeChat deployment, the gap between what users expect and what the license covers will widen. Someone will eventually test whether “as is” holds up in court when an AI agent fabricates data in a board presentation.