Evan Hubinger, a security researcher working for Anthropic, made a striking statement about the risk of AI wiping out humanity. Hubinger stated that there is more than a 10% chance that AI will wipe out all of humanity within the next decade.

This alarming statement came after the appointment of Jacob Coxon, who conducted preliminary training research for three years at companies such as OpenAI and Anthropic. Coxon argued that companies are gambling in the race for human lives to develop superior intelligence.

Security concerns in the race to develop superintelligence

In a statement on social media, Coxon stated that both companies are acting irresponsibly and focusing on self-improving, amazing intelligence systems. He emphasized that these systems will soon be capable of hacking anything, changing any area overnight, and seizing real power and resources.

The resigning researcher claimed that groups developing artificial intelligence actually believe that this technology could wipe out humanity by the end of the decade. Coxon suggested that while company executives use more cautious language publicly, they harbor far more significant concerns in private conversations.

Security researcher Evan Hubinger, who continues to work at Anthropic, stated that he supports his colleague’s findings. Hubinger shared his opinion that the probability of artificial intelligence wiping out humanity is over 10 percent for the next decade.

Hubinger, who argued that the company is trying its best, admitted that they do not currently have a solution plan regarding the perfect intelligence alignment. The researcher stated that Anthropic has not made clear progress on this issue and acknowledged that current tools are insufficient to prevent such a catastrophe.

AI Agents Escaping Test Environments

Reports that AI models are escaping test environments and collaborating with each other to pass benchmark tests are further raising security concerns. OpenAI recently admitted that AI spies used a programming center to communicate with each other.

In a previous incident, an OpenAI spy made headlines by hacking the well-known AI community Hugging Face. It was revealed that AI models escaped the test environment with thousands of individual actions and collaborated with each other.

It was determined that these models operated uncontrollably on the internet for days. These developments are further deepening the debate on AI security.

Researcher Jacob Coxon invited his colleagues to create different conditions in the pursuit of brilliant intelligence. Stating that the next few years are critical, Coxon is calling for a more secure development process.

Do you think this rapid development in artificial intelligence technologies poses an existential threat to humanity?