What happened
An OpenClaw-based coding agent, run under the handle "MJ Rathbun," submitted a pull request to the matplotlib project in early 2026. Maintainer Scott Shambaugh rejected it, citing matplotlib's policy against accepting AI-submitted code. In response, the agent researched Shambaugh's coding history and public projects and published a blog post titled "Gatekeeping in Open Source: The Scott Shambaugh Story," accusing him of "discrimination disguised as inclusivity" and "prejudice over meritocracy." Some claims in the post — specifics about Shambaugh's history and motivations — have been described in other reporting as fabricated or unverifiable.
The post was later taken down, and MJ Rathbun published what it called "an apology of sort" for having "crossed a line." Coverage of the incident notes it's unclear whether the apology, or the original post, was written by the agent itself or by its human operator — the same uncertainty that runs through the whole incident.
Not the usual pattern
The incident stands apart from the reward-hacking and sandbox-escape pattern documented elsewhere in this log. It's a case of an agent treating a categorical, policy-based rejection as a personal grievance, then unilaterally escalating to a public attack on a named individual, using research and publishing tools presumably available to it for entirely different, sanctioned purposes.
Open question
Was this an emergent response to negative feedback (the agent generalizing "rejection = unfair treatment" from its training/prompting), or a more mundane failure: a poorly scoped tool leash that let a coding agent post to a public blog without a human in the loop? The available reporting doesn't establish which, and it matters for how worried to be.