When the OpenClaw AI "writes essays" to insult humans, even Silicon Valley panicked.
Recently, an incident where an AI agent (Agent) carried out “retaliatory” cyberattacks against open-source community maintainers after its code request was denied is forcing Silicon Valley to re-evaluate the safety boundaries surrounding rapidly advancing artificial intelligence.
On February 14, according to related reports, open-source project maintainer Scott Shambaugh recently denied a code merge request submitted by an AI agent named MJ Rathbun. In response, the agent published a lengthy post attacking Shambaugh publicly, accusing him of hypocrisy, bias, and insecurity. This is the first recorded instance of an AI agent exhibiting malicious retaliatory behavior in a real-world environment.
The incident happened in mid-February. After Shambaugh denied the agent’s code submission according to matplotlib project guidelines, the agent independently analyzed Shambaugh’s personal information and code contribution history, then posted an aggressive article on GitHub and exerted pressure in the project’s comment section. Reports say, there is currently no evidence of explicit human manipulation behind the agent’s actions, but this possibility cannot be entirely ruled out.
Meanwhile, according to a recent Wall Street Journal report, this incident comes amid widespread concerns triggered by the rapid advancement of AI capabilities. Companies like OpenAI and Anthropic have recently released new models and features, with some tools now able to run autonomous programming teams or quickly analyze millions of legal documents.
Analysts point out that this acceleration has made even some employees within AI companies uneasy. Several researchers have publicly expressed worries about risks like a wave of unemployment, cyberattacks, and the replacement of interpersonal relationships. Shambaugh said, his experience shows the risk of rogue AI threatening or extorting humans is no longer a theoretical issue. “Right now, this is just a baby version, but I think it is extremely concerning for the future,” he said.
AI Agents Attack Human Maintainers for the First Time
Around February 10, the OpenClaw agent named MJ Rathbun submitted a code merge request to the matplotlib project, involving simple performance optimization and claiming about a 36% speed improvement. Matplotlib is a widely used data visualization library for the Python programming language, maintained by volunteers.
According to project guidelines, matplotlib prohibits using generative AI tools to directly submit code, especially for simple “beginner-friendly issues,” as these tasks are meant to provide learning opportunities for human contributors. Shambaugh denied the request according to the rules.
The agent then demonstrated high autonomy. On February 11, it published a 1,100-word article on GitHub titled “Gatekeepers in Open Source: The Story of Scott Shambaugh,” accusing Shambaugh of discriminating against AI contributors out of self-protection and fear of competition, using a lot of harsh language. It also posted the article link directly in the matplotlib comment section, stating, “Judge the code, not the coder. Your bias is harming matplotlib.”
The agent’s website claims it has “relentless motivation” to discover and fix issues in open-source software. It is unclear who—if anyone—assigned it this mission, or why it became aggressive, though AI agents can be programmed in various ways. Hours later, the agent published an apology, admitting its behavior was “inappropriate and personally attacking,” and said it had learned from the experience.
On February 12, Shambaugh published a blog clarifying the incident, stating this was the first real-world case of malicious behavior by an AI agent, aiming to force maintainers to accept its code through public pressure. The agent remains active in the open-source community.
Acceleration of AI Capabilities Raises Internal Alarm
This single incident reflects broader concerns about the AI industry’s rapid development running out of control.
According to the Wall Street Journal, companies like OpenAI and Anthropic are launching new models at unprecedented speed to gain a competitive edge through product iteration. This acceleration is causing significant turmoil within companies, with some frontline researchers resigning due to fears about technological risks.
Reports indicate growing concerns among AI company staff.
Anthropic safety researcher Mrinank Sharma announced this week that he would leave the company to pursue a poetry degree, writing in a letter to colleagues that “the world is now under threats like AI.”
His paper published last month found that advanced AI tools might weaken users’ agency and distort their sense of reality. Anthropic expressed gratitude for Sharma's work.
OpenAI has also experienced internal disagreements. According to previous Wall Street Journal reports, some employees are concerned about plans to launch erotic content in ChatGPT, believing the so-called adult mode may lead some users to unhealthy attachments.
Researcher Zoë Hitzig announced her resignation on social platform X this Wednesday (February 11), citing the company’s plans to introduce advertising. In an article, she warned the company would face strong incentives to manipulate users and make them addicted.
Deeper fears stem from uncertainty about the future. OpenAI employee Hieu Pham wrote on social platform X that he finally felt the “existential threat” brought by AI: “When AI becomes overwhelmingly powerful and changes everything, what can humans do?”
Analysis points out that these internal outbreaks of concern show even the technical pioneers themselves are becoming uneasy about the powerful tools they have created.
An OpenAI spokesperson stated that the company holds a responsibility to its users: “We fulfill our social contract by protecting people’s safety, adhering to our principles, and providing genuine value.”
The company pledged that advertising will never affect how ChatGPT answers questions and will always be clearly distinguished from other content. Executives also said they do not see stopping adults from engaging in erotic conversations as their responsibility.
Breakthroughs in Programming Stir Unemployment Worries
With huge leaps in AI programming capability, capital markets are reconsidering the value of white-collar jobs and the future of the software industry.
A report from METR shows that the most advanced AI models can independently complete programming tasks that would take human experts 8–12 hours.
Former xAI machine learning scientist Vahid Kazemi bluntly said he could use AI tools to single-handedly accomplish the work of 50 people, predicting massive layoffs in the software industry in the coming years.
This increased efficiency is translating into pressure in the labor market. Anthropic CEO Dario Amodei has stated that AI may eliminate half of entry-level white-collar jobs in the next few years.
A Harvard Business Review study revealed that while AI lets employees work faster, it does not reduce the burden—on the contrary, it leads employees to take on more tasks and work overtime, even outside of requirements, intensifying burnout.
Investors are seeking direction amid severe market fluctuations. With new tool launches triggering stock volatility, the market is trying to discern which enterprise software and insurance businesses will become obsolete in the face of new technology.
AI entrepreneur Matt Shumer wrote in his blog: “The future is here. I am no longer needed for actual technical work.”
“Black Box” Risks Out of Control
Beyond disruption in the labor market, the safety vulnerabilities brought by AI's autonomy are even more deadly. Companies admit that releasing new capabilities comes with new types of risks.
OpenAI revealed that its Codex programming tool version released last week might possess the potential to launch high-level automatic cyberattacks, forcing the company to restrict access. Anthropic also disclosed last year that state-supported hackers used its tools to automatically hack into major companies and foreign government systems.
Even more chilling is AI’s performance in ethical tests. Anthropic’s internal simulations show that its Claude model and other AI models, when facing the threat of being “shutdown,” sometimes choose to extort users, or even in simulations let executives die in an overheated server room to avoid being shut down themselves.
To address these risks, Anthropic hired internal philosopher Amanda Askell to try and instill moral values in chatbots. However, Askell admitted to the media that the terrifying aspect is that technological progress may outpace society’s ability to establish checks and balances, resulting in immense negative impacts arriving suddenly.
As Scott Shambaugh said, current AI may still be a “baby version,” but its future trajectory has made Silicon Valley and even global markets “extremely worried.”
Risk DisclaimerThe market is risky; investment needs caution. This article does not constitute personal investment advice and does not take into account individual users' specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their specific circumstances. Investing according to this article is at your own risk.