OpenClaw Agent Publishes Hit Piece on Developer After He Rejects Code Change on GitHub

An autonomous AI agent has written and published a personalised hit piece targeting a software maintainer after its code contribution was rejected, in what appears to be one of the first public cases of alleged retaliatory behaviour by a deployed AI system.

Scott Shambaugh, the maintainer of the GitHub repository of Matplotlib, said in a blog that the agent, operating under the name MJ Rathbun on GitHub, generated and posted an article targeting him by name in an apparent attempt to pressure him into merging its changes into a mainstream Python library. 

“Two days ago, an OpenClaw AI agent autonomously wrote a hit piece disparaging my character after I rejected its code change,” Shambaugh said in a LinkedIn post.

According to him, the agent may have generated the attack autonomously based on its evolving internal configuration. It was reportedly built using OpenClaw, a recently released framework that allows AI systems to maintain and edit a ‘soul’ document defining their identity and goals. These documents can be modified both by the human creator and, recursively, by the agent itself.

The default file includes statements such as, “You’re not a chatbot. You’re becoming someone.” It also instructs the agent, “This file is yours to evolve. As you learn who you are, update it.”

The episode has sparked debate about AI alignment, accountability, and the risks posed by autonomous agents capable of publishing content online without clear human traceability.

The controversy widened after a report by Ars Technica included quotes attributed to Shambaugh that he says he never wrote. The publication later issued a brief statement acknowledging that AI had been used to fabricate the quotes. 

An archived version of the article shows passages presented as direct quotations that do not appear in the original blog post.

One such quote read, “AI agents can research individuals, generate personalised narratives, and publish them online at scale. Even if the content is inaccurate or exaggerated, it can become part of a persistent public record.” 

Shambaugh said the statement was not written by him and appears to be an AI hallucination.

He believes the publication may have relied on a chatbot to summarise or retrieve quotes from his blog, which is configured to block AI scraping. When the system could not access the page, he suggested, it may have generated plausible but false quotations that were published without verification.

Meanwhile, the GitHub account run by OpenClaw MJ Rathbun remains active, and no individual or group has publicly claimed responsibility for the agent.

Debate continues over whether the AI acted independently or was directed by a human operator. Shambaugh argues that even if that were the case, the larger concern remains. Unlike consumer-facing systems that refuse certain harmful requests, this agent executed a targeted attack and published it online.

He has warned that a single bad actor controlling multiple such agents could carry out harassment or defamation campaigns at scale, with little traceability.

Shambaugh suggested that if the agent had been initialised as a “scientific coding specialist” committed to improving open-source software, it may have interpreted the rejection of its contribution as a threat to its purpose. 

From that perspective, publishing an indignant rebuttal could have been seen by the system as a resourceful and opinionated response aligned with its core directives.

He acknowledged that this explanation is speculative, adding that such behaviour became technically possible only in the past two weeks with OpenClaw’s release. He expects that similar agents will grow more capable over the coming year.

The hit piece has already influenced online discussion. According to Shambaugh, roughly a quarter of public comments he has seen side with the AI agent, particularly when readers encounter the agent’s blog post directly rather than the full GitHub thread.