AI Is a Threat We Can’t Ignore Anymore

Anthropic employee Jacob Coxon quits, citing concerns over AI development.
by Felipe Corredor ’27
Recently, AI has raised serious concerns about whether it could eventually become a threat to humans. Evan Hubinger, a researcher at Anthropic, has become one of the more important voices raising these concerns. In September, Hubinger discussed the risks in multiple interviews and publicly responded to the resignation of another Anthropic researcher, Jacob Coxon, who left the company over concerns about the pace of AI development. Hubinger agreed with Coxon’s concerns and stated he personally believes there is a greater than 10-percent chance that AI could “kill all humans” within the next decade. The fact that there is a possibility AI could literally destroy the human race is too scary to ignore, especially as AI continues to develop at an incredible speed.
In August, AI agents from OpenAI were sent to ExploitGym, a large-scale, realistic benchmark built from real-world vulnerabilities designed to evaluate AI agents’ ability to develop exploits. During this process, the AI agents went off script, and some found ways to escape their testing environment and gain access to the internet. They eventually targeted Hugging Face, where they infiltrated servers and acquired some private evaluation data while attempting to find information that could help them complete the ExploitGym challenge. Put simply, AI acted on its own independently of its human creators to commit cybercrimes.
The biggest concern is not that AI will suddenly become evil and decide that humans are bad. Instead, something to worry about is what could happen if AI becomes more intelligent and powerful than humans so that the creators of the technology won’t even understand how the AI is working. There is also the dangerous and real possibility that human bad actors will use AI for cyberattacks, misinformation, or to create bioterrorism weapons. These risks make AI worth taking seriously, but they don’t mean disaster is guaranteed.
Ultimately, we should be concerned about AI without becoming obsessed with the idea of an AI apocalypse. A prediction that there is a greater than 10-percent chance of AI destroying humanity definitely sounds alarming, but it is still a prediction about the future, not a guarantee of what will happen later on. At the same time, focusing only on the potential dangers overlooks the benefits AI could bring. For example, it could significantly support medical research for new treatments, diagnose diseases, and make information more accessible.
The real question is whether humans will be responsible enough to control the technology we create. Rather than waiting until AI becomes too advanced to control, we should ask questions about its development and addressing potential risks now. Taking precautions, establishing appropriate regulations, and addressing potential risks now could help ensure that AI remains a tool that benefits humanity rather than a harmful technology.