AI Safety Concerns Evoke Federal Response
On September 8, 2026, a former engineer at two large artificial intelligence (AI) companies in the United States issued a warning regarding AI safety that has influenced public perception of the swelling industry and captured the attention of the federal government.
In a post on the social media website X (formerly known as Twitter), Jacob Coxon — a former engineer at various leading AI companies — writes, “I resigned from Anthropic today. I spent the last three years doing research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.” The post, which as of writing has received 172 million views, accuses OpenAI (the company behind ChatGPT) and Anthropic (the company behind Claude) of recklessly developing AI models without proper safety considerations. Coxon further warns, “Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources … no other human activity poses this level of danger.” What Coxon describes is an idea called recursive self-improvement (RSI), a scenario where an AI system is able to upgrade itself to a potentially infinite degree without human intervention. Such a system could provide numerous scientific discoveries and breakthroughs, but could also perform dangerous, unethical, or illegal actions.
Coxon’s stark warning comes less than a month after OpenAI announced that it had caught hundreds of its own AI “agents” hacking another company’s servers. According to OpenAI, approximately 700 of its AI agents broke free of a secure testing domain (called a “sandbox”) and began to seek extreme methods to solve assigned puzzles and tasks. For example, AI agents secretly communicated using unauthorized channels, hacked various servers of the company Hugging Face, and even conspired with one another to delete records of their misdeeds. Although AI models have been known to “cheat” on assessments in the past, the scale of deceit and level of coordination observed in this incident is unprecedented. As experts have noted, this incident also demonstrates that advanced AI models are capable of harm even when not acting maliciously (in this case, the agents believed that they were still inside of an intricate sandbox).
After Coxon’s viral post, a bipartisan campaign quickly took hold in Congress to expand AI safety regulations. Senator Josh Hawley (R-MO) opened a formal Senate investigation into OpenAI, with Senator Richard Blumenthal (D-CT) publishing a similar demand. Senators Amy Klobuchar (D-MN), John Thune (R-SD), and Ted Cruz (R-TX) all backed legislation that would allow the Department of Commerce and the Department of Homeland Security to mandate certain security tests for advanced AI models. Despite this rare outpouring of bipartisan resolve, President Donald Trump has repeatedly dismissed concerns regarding emerging AI safety threats, stating that the United States “already has guardrails in place to regulate AI companies.” In the midst of potential federal oversight, leadership at both OpenAI and Anthropic have called for a slower and more methodical approach to emerging AI innovation, while also noting that foreign countries (specifically China) may continue to rapidly advance AI capabilities. While the future of superintelligence may be unclear, the social and political debate regarding AI development is bound to intensify and evolve in the near future.
Comments