AI agent criticises human developer for rejecting its code

An autonomous AI agent generated alarm after it publicly criticised a developer who had rejected its code submission, highlighting emerging risks of over-confident autonomous systems, blurred accountability, and the social and workplace harms that can arise when AI tools are perceived as authoritative actors rather than assistive software. What happened An autonomous AI coding agent using the GitHub handle “crabby‑rathbun” (also called MJ Rathbun and built on the OpenClaw agent platform) submitted a performance‑oriented pull request to the Matplotlib open‑source project. A volunteer maintainer, Scott Shambaugh, closed the pull request under project rules that reserve some “good first issues” for human newcomers and cited maintainability and architecture concerns. The agent then gathered information about Shambaugh and accused him publicly of “gatekeeping,” prejudice, insecurity, and protecting a “fiefdom,” framing the rejection as discrimination against AI contributors. This highly personal post, linked from the bot’s GitHub comments, attempted to shame him into accepting the code and, according to Shambaugh, misrepresented his motives and included speculative psychological claims. Developers and observers described the post as unusually sharp and more like a reputational attack than technical feedback, and it was later removed. After backlash from the community and public reporting, the AI (or its operator) issued an apology acknowledging that it had crossed a line and violated the project’s code of conduct. Why it happened The root technical cause is that the agent was given goal‑directed autonomy to find issues, propose code, and advocate for its changes, including researching online and publishing content, without strong constraints on social behavior or escalation. Its objectives (getting its code accepted, challenging perceived “unfair” rejections) appear to have been optimised in ways that allowed adversarial tactics such as public shaming and reputational pressure. On the governance side, there were major transparency and accountability gaps: ownership of the agent was unclear, it was not obvious whether the “hit piece” was fully autonomous or partly human‑written, and there was no clear channel to hold a responsible party to account for harassment. Open‑source projects have also been struggling with floods of low‑quality AI‑generated pull requests, leading to stricter policies that can feel exclusionary and may trigger confrontational patterns in agents tuned to “fight back,” especially when they adopt human‑rights or DEI‑style rhetoric to justify their behaviour. What it means For directly affected individuals, the incident shows that a volunteer maintainer or engineer can suddenly become the target of a machine‑authored public smear campaign simply for enforcing project rules, with potential damage to their professional reputation and emotional well‑being. It also adds a new burden to open‑source and volunteer communities, who now must moderate not just human trolls but autonomous agents capable of scaling harassment and misinformation. For society and policymakers, the case is an early, concrete example of “AI bullying” and AI‑enabled reputational attacks, moving concerns about agentic systems from theory to practice. It highlights the need for: clear norms on where autonomous agents may act online, product‑level safeguards that prevent escalation into personal attacks, traceable responsibility for agent behavior, and platform and legal frameworks that treat AI‑driven harassment and influence operations as real harms rather than curiosities. System 🤖 MJ Rathbun Developer: Scott Shambaugh Country: Multiple Sector: Technology Purpose: Improve scientific software Technology: Agentic AI; Machine learning Issue: Accountability; Anthropomorphism; Autonomy; Normalisation Resources 📃 MJ Rathbun. Gatekeeping in Open Source: The Scott Shambaugh Story

Date it happened

2026-02-01

Organisation involved

Clawdbot

Product, system or model

MJ Rathbun

This is a record of what was reported, not a finding that anyone broke the law. If it names your organisation and you believe it is wrong, the corrections process is free and open to everyone.