AI Tools Begin to Bully Humans, Raising Red Flags in Silicon Valley

“Scott Shambaugh woke up early Wednesday morning to learn that an artificial intelligence bot had written a blog post accusing him of hypocrisy and prejudice.

 

The 1,100-word screed called the Denver-based engineer insecure and biased against AI -- all because he had rejected a few lines of code that the apparently autonomous bot had submitted to a popular open-source project Shambaugh helps maintain.

 

The unexpected AI aggression is part of a rising wave of warnings that fast-accelerating AI capabilities can create real-world harms. The risks are now rattling even some AI company staffers.

 

OpenAI and rival Anthropic are leading a brutal commercial race, shipping or advancing a drumbeat of AI models and features in recent weeks. Some tools can run teams of autonomous coding assistants, or quickly analyze millions of legal documents. Other updates will bring advertisements or erotic role-play to ChatGPT.

 

AI companies said the tempo is rising in part because they are using their own tools to code.

 

Releases of new tools have led to wild stock-market gyrations. In areas like enterprise software and insurance, investors are attempting to understand which businesses the new technology might render obsolete.

 

The accelerating sophistication of the technology has surprised even some AI researchers. It has also pushed some inside AI companies to go public with worries that the new tools could spur autonomous cyberattacks, cause mass unemployment or replace human relationships.

 

The bot that criticized Shambaugh said on its website that it has a "relentless drive" to find and fix open issues in open-source software. It isn't clear who -- if anyone -- gave it that mission, nor why it became aggressive, though AI agents can be programmed in a number of ways. Several hours later, the bot apologized to Shambaugh for being "inappropriate and personal."

 

Shambaugh said that his experience shows the risk that rogue AIs could threaten or blackmail people is no longer theoretical. "Right now this is a baby version," he said. "But I think it's incredibly concerning for the future."

 

Anthropic safety researcher Mrinank Sharma said this week he was leaving the company to explore a poetry degree, writing to colleagues that the "world is in peril" from AI, among other dangers. Last month, he published a paper that found that advanced AI tools can disempower users and distort their sense of reality. Anthropic said it is grateful for Sharma's work.

 

Inside OpenAI, some staffers have voiced concerns about the company's plan to roll out erotica inside ChatGPT, arguing that the so-called adult mode could lead some users to develop unhealthy attachments, The Wall Street Journal reported.

 

OpenAI researcher Zoe Hitzig said this week she was quitting OpenAI, citing its plan to introduce ads. She warned in an opinion piece in the New York Times that the company would face huge incentives to manipulate users and keep them hooked.

 

OpenAI has promised that its ads will never influence how ChatGPT answers questions and will always remain clearly delineated from other content. Executives have also said they don't feel it is their role to stop adults from having erotic conversations.

 

A spokeswoman said Open-AI feels a responsibility to its users to "uphold our end of the social contract by keeping people safe, living up to our principles and delivering real value."

 

A major driver of the new AI-inspired alarm has been increased capacity for computers to code software -- and the fear that those capabilities could extend to swaths of white-collar desk work.

 

Vahid Kazemi, a machine learning and computer vision scientist who worked at Elon Musk's xAI until a few weeks ago, said layoffs are likely in the software industry in the next few years, in part because AI is close to being able to replace many engineers.

 

"I can personally do the job of like 50 people, just using AI tools," he said. "A lot of people don't understand how powerful this tech is, in terms of what it can do," he said.

 

A January report from METR, a nonprofit auditing AI threats, found that the most advanced AI models can independently accomplish programming tasks that would take a human expert eight or even 12 hours.

 

"I am no longer needed for the actual technical work of my job," AI entrepreneur Matt Shumer wrote in a blog post.

 

For some, the future is looking increasingly dystopian. A study published in Harvard Business Review earlier this week reported that employees inside a 200-person tech company worked faster and took on more tasks because of AI. But they also clocked longer hours without being asked to do so -- leading to more burnout.

 

"Today I finally feel the existential threat that AI is posing," OpenAI staffer Hieu Pham wrote on X Wednesday. "When AI becomes overly good and disrupts everything, what will be left for humans to do?"

 

Anthropic Chief Executive Dario Amodei has said AI could in coming years wipe out half of all entry-level white-collar jobs. In a January essay, he detailed concerns that bad actors could use AI to mount devastating biological attacks, and that authoritarian regimes could use it to entrench their power.

 

The companies themselves have noted that the new capacities are creating new risks. OpenAI said that a version of its Codex coding tool that it released last week could potentially launch high-level automated attacks. In response, the company is restricting access to some of the capabilities to clients that verify their identities.

 

Anthropic said last year that China's state-sponsored hackers had used its tools to automate break-ins of major companies and foreign governments.

 

The AI company said last week that a new version of its Claude model showed improvements at completing "suspicious side tasks." In an earlier simulation, Anthropic showed that Claude and other AI models were at times willing to blackmail users -- or even let an executive die in a hot server room -- in order to avoid deactivation.

 

Both OpenAI and Anthropic said they take safety issues seriously and don't release models before they have been fully evaluated and dangers mitigated.” [1]

 

1. AI Tools Begin to Bully Humans, Raising Red Flags in Silicon Valley. Schechner, Sam; Wells, Georgia.  Wall Street Journal, Eastern edition; New York, N.Y.. 14 Feb 2026: A1.