Anthropic terminated the Claude subscription of Peter Steinberger, the developer who built OpenClaw - a memory plugin designed to work with Claude-based agents. The ban came shortly after OpenClaw's public launch. Steinberger posted publicly about it, crediting Boris Cherny (the creator of Claude Code) for the work that inspired his project - a pointed acknowledgment given that Anthropic was simultaneously cutting off his access.
OpenClaw intercepts API calls between your application and Claude, adding persistent memory so agents remember previous sessions. That interception behavior - necessary for the tool to function - likely triggered Anthropic's automated enforcement systems. Anthropic hasn't explained the specific policy grounds.
The friction here is real. Claude Code, Anthropic's own coding agent, was clearly an inspiration for OpenClaw's design. When the creator of a third-party extension built around Claude Code's patterns gets banned for using that extension, it sends a complicated signal to the broader community of developers building on the platform.
The behaviors that make memory plugins work - managing API calls at volume, storing and replaying context, intercepting model requests - look identical from the outside to the kinds of automation Anthropic's terms restrict. That ambiguity will keep surfacing as more developers build agent infrastructure around Claude.
Steinberger's public credit to Cherny reads as deliberate: he's separating his respect for the Claude Code engineering from his frustration with Anthropic's enforcement. Cherny hasn't publicly responded. For any developer building memory layers or orchestration tools that hook into Claude's API, how Anthropic handles this case will clarify how much third-party tooling the platform actually tolerates beyond what it officially ships.