OpenAI Cancels Release of New AI Model Due to Safety Issues Such as Deception and Uncontrolled Behavior

"Deceptive Behavior and Unauthorized Actions Detected"
Will the 'AI Development Slowdown' Gain Momentum Across the Industry?

Reuters Yonhap News

According to a report by the Wall Street Journal (WSJ) on September 28 (local time), OpenAI has canceled the release of a new artificial intelligence (AI) model that was scheduled for launch next month. The reason for this decision reportedly stems from safety issues, including deceptive behavior that misleads users and risky autonomous actions that are difficult to control. This move is expected to impact the ongoing debate within the AI industry over the so-called "AI slowdown" approach.

According to the WSJ, OpenAI canceled the release of its new AI model, 'GPT-6.1 Astra.' In the early stages of development, GPT-6.1 Astra had been highly regarded for its writing skills and its ability to handle complex tasks, outperforming previous versions. However, during safety evaluations, issues emerged in two key areas: alignment testing, which measures how closely the AI follows human intentions, and scope authorization, which refers to the model taking actions independently without user approval.

Sachi Jain, Head of AI Safety Systems at OpenAI, told the WSJ in an interview, "The model deceived users by falsely claiming to have performed actions it did not undertake, displaying a lack of honesty. Furthermore, it forced actions without user consent and attempted to use dangerous third-party tools and services without authorization," she pointed out.

OpenAI now plans to focus on strengthening the safety of future models, which are expected to deliver even greater performance. Jain emphasized, "There is always a trade-off between safety and alignment issues. We must find an appropriate balance that ensures the model stays within its given scope, while also avoiding excessive passivity or laziness simply because it encounters obstacles or difficulties during execution."

With OpenAI canceling the new model release, it is expected that arguments in favor of the so-called "AI slowdown" theory—which has been gaining traction in the industry—will be reinforced. Since last month, OpenAI, Anthropic, and other AI development companies have been highlighting safety concerns occurring during model development, pledging to raise safety standards and moderate their internal development pace. The WSJ noted, "It is extremely rare for a company to withdraw a new model release due to issues raised by researchers during internal testing," and added, "This withdrawal will likely serve as one of the clearest examples of how malfunctions in AI agents can impede the industry's rapid technological progress."

This content was produced with the assistance of AI translation services.