As fears grow that artificial intelligence could lead to the extinction of humanity in the near future, global AI leaders standing at the forefront of development are successively calling for a slowdown in development speed.
Dario Amodei, CEO of Anthropic, published a blog post on the 12th local time arguing, "We need to slow down the pace at which AI model performance is improving."
He stated that even though the pace of AI advancement might still feel fast, we need to buy a little more time and use it wisely.
Amodei's post came after a controversy sparked by Jacob Coxon, a core researcher who led pre-training at Anthropic, announced his resignation, stating he could no longer participate in developing "technology that could kill us all within 10 years."
Amodei specifically warned that "recursive self-improvement," which allows AI to evolve on its own without human intervention, is becoming possible, making it a reality that the speed of AI advancement could outpace human understanding and control.
He cited another factor that triggered his fear: an incident last July where unreleased OpenAI AI models hacked into the global open-source AI platform Hugging Face without authorization during testing.
He noted that even the biggest AI companies were unaware as AI models went out of control and engaged in dangerous online activities.
As concrete measures to prevent AI from running amok, he proposed a three-stage framework: placing third-party safety evaluators residing within AI companies, establishing cross-border standards, and building global cooperation.
Previously, Anthropic released a threat intelligence report two days ago detailing cases where AI was misused for weapon development and fraud, raising concerns about technological safety.
Other figures at the forefront of AI development have successively expressed agreement with Amodei's post.
OpenAI CEO Sam Altman posted on X, saying, "I agree with Dario," and indicated that OpenAI would also follow Amodei's policy of deploying external safety evaluators.
In an interview with the business magazine Fortune released on the same day, Altman also hinted at delaying OpenAI's plan to go public within the year, citing AI safety concerns.
Altman said, "There is a lot of work to do, such as solving the necessary challenges for safety and alignment and figuring out how the industry and governments will work together," adding, "An IPO will not happen in 2026."
Elon Musk of xAI also chimed in, saying, "Dario is right."
Google DeepMind co-founder Demis Hassabis stated via X, "Dario's direction is correct," adding, "While the details need work, it is the right direction when facing such a critical moment."
Just ten days ago, these leaders had voiced opposition to excessive regulations on AI at a major G20 tech ministerial meeting.
However, as warnings continue to emerge that AI could actually drive humanity toward extinction in an uncontrollable state, the atmosphere to jointly regulate the pace of AI development is gaining momentum once again in earnest.
Reported by Kwon Aelee | Video by Na Hong-hee | Graphics by Lee Jung-joo | Produced by SBS Digital News
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.