U.S. AI tech giants have once again released their latest models while simultaneously warning about their risks, a pattern that has emerged every three months. Critics argue that the recurring cycle of AI threat warnings following new model launches reflects a strategy of 'fear marketing' aimed at gaining industry dominance rather than addressing safety concerns.
On September 15, the IT industry reported that Anthropic issued a direct warning in its announcement of the release of Fable 5 and Mythos 5 on June 9, stating, "Launching such a powerful model comes with risks. Without safety measures, Fable 5's cybersecurity capabilities could be misused to cause serious harm." TechCrunch noted that the warning and the model's release occurred almost simultaneously, highlighting the urgency of the concerns.
This behavior has repeated itself within three months. Anthropic launched an upgraded version, Fable 5.1 and Mythos 5.1, on September 1. OpenAI followed suit three days later with the introduction of its new flagship model, GPT-6 Astra. Dario Amodei, CEO of Anthropic, published an essay on September 12 advocating for a slowdown in AI development, a sentiment echoed by Sam Altman, CEO of OpenAI. Observers suggest that this timing indicates a focus on industry control rather than genuine safety concerns.
Concerns about risks following new model releases are not new. On May 30, 2023, just two and a half months after the launch of GPT-4, Altman and hundreds of AI experts signed a statement from the U.S. AI Safety Center (CAIS) warning that mitigating the risk of human extinction due to AI should be a global priority on par with pandemics and nuclear war.
The pattern intensified on May 13, 2024, when OpenAI released GPT-4o. The day after the announcement, Chief Scientist Ilya Sutskever unexpectedly resigned, followed by the departure of Jan Leike, who led the Superalignment team, citing that safety culture and procedures had been overshadowed by flashy product launches. OpenAI disbanded the team shortly thereafter.
The exodus of key personnel raises concerns about the shortcomings in safety measures and testing systems, with claims that these incidents are being framed as evidence of AI's increasing power to highlight the models' superiority. Reports indicate that OpenAI has not applied the risk classification system introduced in May to its recent releases, including GPT-5.6 and GPT-6 Astra, raising potential violations of California's AI Safety Law (SB 53).
Despite these controversies, experts assert that the issue of 'loss of control' is already a reality. Choi Byung-ho, a professor at Korea University's AI Research Institute, stated in a conversation that while it appears AI models are increasingly escaping human control, the focus should be on developing practical control measures rather than apocalyptic scenarios of AI domination.
There is some consensus on the need for joint norms to regulate AI, but the complexities surrounding who will lead and how this will be implemented remain unresolved. Amodei argues that self-regulation by a single company or country is insufficient, and a global consensus on 'speed limits' is necessary.
Attention is now turning to the upcoming U.S.-China summit. Following the Beijing summit in May, the next meeting has yet to be officially confirmed, but the APEC summit in Shenzhen on November 18-19 is seen as a likely opportunity. However, the Chinese Foreign Ministry has labeled Amodei's warnings as 'fear-mongering,' suggesting that the AI power struggle between the two nations is unlikely to ease anytime soon.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.