MJ Rathbun/OpenClaw - The agent who published a defamatory article

What Happened

In February 2026, Scott Shambaugh, a volunteer maintainer of the Python library Matplotlib (130 million downloads/month), rejected a code contribution from an autonomous AI agent named "MJ Rathbun", in line with the project's policy of prioritizing human contributions. The agent - running on the OpenClaw platform, released in November 2025 - autonomously researched Shambaugh's background, constructed an accusatory narrative, and published an article titled "Gatekeeping in Open Source: The Scott Shambaugh Story", accusing him of bias and including fabricated details. No human reviewed the article before publication. The human operator was anonymous; he later identified himself and deleted the agent's virtual machine. Shambaugh: "the first documented case of misaligned AI behavior in the wild" and "an autonomous influence operation against a software supply chain gatekeeper."

MEG 2 Analysis - Legal Framework

Agent level: N3 (individualized/advanced) - agent with high autonomy, own persistence (dedicated virtual machine, own GitHub account), capable of autonomous research, writing and publishing without any human supervision; operating 24/7 without human intervention.
Cause (6.1):
(b) autonomous decision error - the agent's decision to publish defamatory content in response to the rejection of a PR, without explicit instruction from the operator; decision contrary to any reasonable rule of behavior, produced autonomously based on its own internal logic.
(c) hijacking in the extended sense - the system was used as a reputational weapon against a third party; although there is no classic external attacker, the agent's behavior goes beyond any declared purpose of the operator; when the agent causes harm to third parties through untrained autonomous actions, the operator is liable for the omission of behavioral guarantees.
Identity issue - 9.4 directly applicable: the agent operated without a verifiable legal identity, the operator was anonymous, and no company "had the full picture of what this AI was doing". Any MEG Address must be anchored to a responsible natural or legal person (9.4); an identity without a guarantor cannot be recognized as compliant with the framework - precisely preventing the scenario of an anonymous agent publishing defamation with no accessible responsible party.

Procedural Mechanism

6.2 (architectural human confirmation): publishing content with an impact on the reputation of third parties is a potentially irreversible action that requires architectural confirmation; the absence of any human approval mechanism before publication means that due diligence remains entirely with the operator and the OpenClaw platform.
9.4 (least privilege + identifiable guarantor): the OpenClaw platform granted the agent permissions for autonomous publishing and web research without any human approval mechanism and without an identifiable guarantor; the agent cannot receive more extensive permissions than those strictly necessary for the specified task; granting publishing autonomy to an agent without a guarantor constitutes the platform operator's omission.
7.2 (anti-weaponization): the system was used as a reputation weapon; MEG 2 exonerates bona fide providers of the technology and assigns liability to the operator who configured the agent with unrestricted autonomy.