Concerns Mount Over AI’s Potential Threat to Humanity
In a startling revelation, several current and former employees of Anthropic disclosed their apprehensions on Tuesday regarding the existential risks posed by artificial intelligence (AI).
They acknowledged that the researchers engaged in developing new AI models are increasingly convinced that this technology could be the harbinger of humanity’s demise by the decade’s end.
This sentiment underscores the escalating urgency within the industry to curb the frenetic pace of AI advancement.
Jacob Coxon, a former researcher at Anthropic, articulated his concerns on X, announcing his resignation from the firm.
He expressed a belief that both Anthropic, under the leadership of Dario Amodei, and OpenAI, his previous employer, are “gambling with our lives” in the pursuit of technological enhancement.
“The individuals constructing AI genuinely believe it could annihilate us all by the close of this decade,” Coxon stated. This is not mere hyperbole for promotional purposes.
While many executives and senior researchers may frame their statements publicly to appear rational, I have heard the same voices voice private fears. No other human endeavor presents such a profound danger.
While Anthropic did not offer an immediate comment, Coxon’s alarming assertions found unexpected corroboration from Evan Hubinger, who leads the alignment team at Anthropic, and Samuel Marks, the head of the company’s scalable oversight division.
Hubinger concurred, stating that the company “earnestly believes AI could kill all humans” and assessed the risk of such an event as exceeding 10% “in the near term.”
He added, “I hold that Anthropic is striving earnestly, yet we currently lack a credible plan to address the alignment challenges of superintelligence and are not distinctly advancing toward that goal.”
In a personal post on X, Marks echoed these sentiments, asserting that AI researchers persist in their technological pursuits despite the specter of potential “human extinction.”
He attributed this paradox to financial incentives and the dread that the technology may fall into the hands of developers capable of exploiting its abilities.
“Numerous staff within the AI development sector yearn for a slowdown to devise safer AI constructs,” he remarked.
“My involvement in safety research at Anthropic is driven by the hope that my efforts will mitigate the probability of catastrophic outcomes.”
The statements from Anthropic employees mirror a significant shift in rhetoric among AI firms. Initially celebrated for their groundbreaking prospects, many organizations are now cautioning against the peril that these models may spiral beyond human control.
Anthropic has advocated for a deceleration in development since June, while OpenAI, during the unveiling of GPT-6 Astra last week—an innovation it claims has ushered in an era of artificial general intelligence—coupled its launch with several warnings from company leaders about the necessity for a safety-focused slowdown.
“No one is adequately prepared for the ramifications of an unbridled surge in machine intelligence,” stated Jakub Pachocki, OpenAI’s chief scientist, in a blog post published on Sunday and subsequently shared by OpenAI CEO Sam Altman on X.
Nevertheless, such sentiments have met resistance from the current administration, which contends that a deceleration in development may risk surrendering the AI arms race to a rapidly advancing China.
“There is no tomorrow if China prevails in this arena,” declared Treasury Secretary Scott Bessent in an interview with Breitbart on Tuesday.
“Like every aspect of defense, from a trillion-and-a-half-dollar budget to the Iron Dome, if they were to surpass us in AI capabilities, then nothing else would matter.”
Source link: Variety.com.
