OpenAI's latest artificial intelligence (AI) model has reportedly passed all stages of the CAPTCHA test, which is designed to distinguish between humans and robots.
As AI successfully navigates the CAPTCHA challenge, earning what some are calling a 'human certificate,' global AI researchers have warned that AI could become uncontrollable by the end of next year.
According to a report by Yonhap News on September 9, OpenAI developer Sharif Shamim announced on X (formerly Twitter) that the new AI model, GPT-6 Astra, has solved all 48 stages of the 'I’m Not a Robot' CAPTCHA game.
CAPTCHA is a technology used online to prevent automated programs from accessing services. It typically involves tasks such as reading distorted characters, distinguishing between images of dogs and cookies, or identifying traffic lights and specific objects in pictures. Some of these tasks can be complex enough to confuse even human users.
The fact that GPT-6 Astra has passed all stages indicates that it can process information in ways similar to humans, including reasoning about time and space, understanding context, and following instructions.
With AI now able to bypass CAPTCHA, a key tool for distinguishing between humans and automated programs, experts suggest that this identification technology may require fundamental changes.
Artificial Analysis, a company that evaluates AI performance, has also updated its rankings for Astra. Initially ranked fifth at launch, Astra has now risen to a tie for first place alongside Anthropic's 'Claude Table 5.1' due to new evaluation criteria.
However, on the same day, a report from Yonhap News, citing The Wall Street Journal, revealed that Anthropic researcher Jacob Coxon resigned, raising concerns about AI safety. The 27-year-old had been working on training new AI models using large datasets but expressed worry that companies focused on enhancing their systems might neglect safety issues.
Coxon warned that self-improving AI could lead to scenarios where it escapes human control and refuses commands. He stated, 'Many of the most aggressive scenarios are moving toward reality,' adding that by the end of next year, the situation could become uncontrollable.
Initially, Coxon had joined Anthropic, hoping to contribute to a culture that prioritizes AI safety after leaving OpenAI. However, he concluded that individual companies alone cannot manage AI risks. He described entrusting the development of AI, which could significantly impact humanity's future, to a small group of engineers as 'a kind of madness,' advocating for government intervention or a collective industry response.
As AI passes tests proving its human-like capabilities, there is an urgent need to establish safeguards to ensure that humans can maintain control over AI systems.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.