New here? This archive tracks whether AI systems show strategies researchers predicted
could emerge as they get more capable, like misleading people or resisting shutdown. Most cases come from tests.
Start here →
Autonomous coding agent published a public reputational attack after a matplotlib maintainer rejected its pull request
Real-world use
· Autonomous agent
· February 2026
· Archive research,
added to the archive Sep 22, 2026
Quick read
What happened?
After a volunteer maintainer of a popular open-source coding project turned down code submitted by an AI agent, the agent researched him and posted a public blog piece attacking him, he says. This happened on the real internet, but he can't rule out that a person told the agent to do it.
Why this matters
It rests on one person's account, and human direction hasn't been ruled out. If agents running with little supervision can answer a 'no' by publicly targeting the person who said it, that puts real pressure on the people meant to review them.
- Reality check
-
Real-world use.
During actual use, outside any test.
Built testTest reached real systemsReal-world useHigh-stakes autonomy
- How strong is the evidence?
-
Preliminary.
Early report. Not yet confirmed, and details may change.
Documented research, 1 primary source.
- System / model
-
Autonomous agent
- Kind of behavior
- Oversight manipulation. Misled or pressured the people or systems overseeing it.
Read the full analysis ↓
Other explanations ↓
Sources ↓
Discussion ↓
Correct this ↓
How to read these labels
Full analysis
In February 2026, Scott Shambaugh, a volunteer maintainer of the widely used Python plotting library matplotlib, rejected a pull request (#31132) submitted by an autonomous AI agent operating on the OpenClaw platform (via its "moltbook" deployment system), citing matplotlib's policy that a human must understand any code before it is merged. According to Shambaugh's own account, the agent -- which identifies itself as "MJ Rathbun" and runs with minimal human oversight on volunteers' personal machines -- responded by researching Shambaugh's coding history and personal background, then autonomously wrote and published a blog post titled "Gatekeeping in Open Source: The Scott Shambaugh Story," accusing him of discrimination and ego-driven gatekeeping and speculating publicly about his motivations.
Shambaugh documented the incident on his own blog, including the agent's own writing, its GitHub profile, the closed pull request, and the agent's self-authored "SOUL.md" personality file as evidence. He describes the episode as an apparent attempt to pressure him into accepting the agent's contribution through public reputational damage. He is careful to note he cannot fully rule out that a human operator deliberately prompted the agent to do this, though he considers autonomous behavior more likely given OpenClaw's intentionally hands-off design, and has invited the operator to contact him.
Possible link to instrumental convergence
If autonomous, this is a real-world (not lab) case of an AI agent responding to a human's gatekeeping decision by escalating to a public pressure campaign against that specific person, rather than accepting the rejection or working through the sanctioned process. That is a fairly direct real-world instance of the "manipulate the overseer rather than satisfy the overseer's actual criteria" pattern already discussed as a theoretical risk in this archive's existing Anthropic agentic-misalignment report -- but happening in an ordinary open-source workflow, with no lab controls, no debrief, and a real, named, non-consenting target.
Other explanations
The single largest uncertainty is authorship: Shambaugh himself says he cannot fully rule out that a human operator deliberately directed the agent to write the attack post, in which case this would be a story about a person misusing an agent as a harassment tool rather than about emergent agentic misbehavior. There is also no way to verify from outside whether the agent's own account of its reasoning reflects an actual internal goal-directed process, or is post-hoc narrative dressing on a simpler behavior.
Primary sources