Anthropic researcher Evan Hubinger has issued a warning that artificial intelligence (AI) could have over 10% chance of “killing all humans” within the upcoming decade, underscoring increasing concerns regarding the risks posed by the cutting-edge AI-powered systems.
The comments follow Jacob Coxon, another Anthropic researcher, announcing his resignation from the company.
Coxon accused Anthropic and the ChatGPT manufacturer of “gambling with our lives,” by rapidly adopting the superintelligence without sufficient security measures.
He further issued a warning that future AI systems could become superhuman, which is possibly able to hack systems, revolutionise industries and acquire significant power.
He stated many people working in AI believe the technology could endanger humanity.
Taking to X (formerly Twitter), Hubinger, an alignment science lead at Anthropic, agreed with Coxon’s assessment, stating, “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
Hubinger said the company does not yet have a solution to align future superintelligent systems with human interests and is not clearly on track to develop one.
Hubinger stressed that the risks posed by current AI models remain reduced. His primary concern is the potential of recursive self-enhancement, in which AI-powered systems could rapidly enhance themselves and assist in generating increasingly capable successors with limited human intervention.
Previously, the Claude AI manufacturer issued a warning that recursive self-improvement could make it harder for humans to feel protected, monitor, and manipulate AI systems.
Concerns about AI becoming uncontrollable have also been raised by technology leaders and researchers worldwide. Recent incidents involving AI systems, including an alleged breach of Hugging Face, have further intensified debate over AI safety.
Concerns regarding AI becoming
Coxon argued that stronger cooperation between AI-powered companies may be necessary to prevent a race toward increasingly powerful systems globally, possibly including some restrictions to enhance model capabilities.
