'I'm Not a Robot' (OpenAI Astra)..... The Boundary Between Robots and Humans Has Become Even More Blurred

[Financial News] OpenAI's new artificial intelligence (AI) model, 'GPT-6 Astra,' has passed all CAPTCHA games used to prove human identity on the internet.
OpenAI developer Sharif Shamim announced on the 8th (local time) via the social networking service X (formerly Twitter) that he had tested GPT-6 Astra to solve the CAPTCHA game "I'm Not a Robot" and that it had obtained a "human certificate" by solving all 48 stages of the game.
'I'm Not a Robot' consists of games such as matching distorted alphabets, distinguishing between cookies and dogs in similar-looking photos, and identifying the location of specific structures like traffic lights within a picture.
Most of these are games currently used in practice to filter out crawling bots and similar devices that scrape websites on the internet. Some are so difficult that even real humans often get the answers wrong.
Astra's success in solving all CAPTCHA games is attributed to its advancement to a level comparable to humans in various domains, including spatiotemporal reasoning abilities such as image recognition, contextual reasoning abilities such as understanding instructions, and computer-use skills.
As commercial AI models render the tools used to distinguish between humans and robots ineffective, a major overhaul of robot-blocking technology is now required.
Artificial Analysis, an AI model performance analysis company, improved Astra's ranking by reorganizing its AAII intelligence-metric evaluation system twice over four days. On the 4th, the day of its launch, Astra entered the rankings in a tie for 5th place with 61 points under the existing evaluation system (version 4.1), falling behind Anthropic's top-ranked 'Claude Fable 5.1' (66 points); however, after the new system (version 4.3) was applied, it scored 53 points and became tied for 1st place with Fable 5.1.
Artificial Analysis explained that it had applied some features of the upcoming '5th edition' ahead of schedule to keep pace with advances in cutting-edge AI technology.
Meanwhile, on the same day, OpenAI announced that it had solved the Navier–Stokes equations, fluid-dynamics equations regarded as one of the world's seven greatest mathematical challenges and one of the Millennium Prize Problems, in 88 hours.
The New York Times (NYT) and others reported this, calling it the most dramatic sign that OpenAI's undisclosed AI model is fundamentally changing the field of advanced mathematics. OpenAI announced that the model completed 165 pages of proofs and verification after mobilizing approximately 10,000 AI agents and consuming 130 billion tokens over 88 hours of computation.

[email protected] Lee Seok-woo, International Specialist Reporter