What might surprise many is the new reality of artificial intelligence. One of the biggest tests to check whether a real human is checking a website on the internet could be the captcha tests. In a recent incident, an AI model has completed all 48 levels of a browser game and beaten this test.
Sharif Shameem, a developer, said on X that GPT-6 Astra successfully completed Neal Agarwal’s ‘I AM Not a Robot’ game. According to Shameem’s post, the AI model progressed through the entire challenge and reached the final screen, where it received a “Human Verification” result.
The demonstration does not mean that AI can defeat every Captcha used by websites. However, it shows that modern AI models are becoming capable of handling interactive computer tasks rather than simply answering basic questions.
How Astra Handled The Captcha Game
Captchas are commonly used by websites to separate human users from automated software. They can ask people to identify objects in images, solve puzzles or perform actions that are intended to be difficult for bots.
Google’s reCAPTCHA is one of the most widely known examples, although several other companies also provide similar anti-bot technologies.
The game created by Agarwal uses a series of different challenges. Players have to identify visual elements, move objects, follow instructions and understand what each level is asking them to do. Completing the full game can take a human player around 30 to 60 minutes.
GPT-6 Astra reportedly used visual understanding together with its Computer Use capabilities to work through the challenges. It could examine the browser interface, understand the objective and perform actions such as clicking, dragging and typing.
As per the Shameem’s post, The test was not limited to simple image recognition. Some levels required the model to understand the rules of a particular puzzle and determine the correct way to complete it.
After completing the final challenge, the game displayed its “Human Verification” result and generated a Verified Human certificate.
Interestingly, OpenAI Head of Applied Research Boris Power also commented on Shameem’s post, saying, “True AGI can go everything a human can do on the computer.” The test therefore offers another example of the progress being made in AI computer-use capabilities.
-Times now
Sharif Shameem, a developer, said on X that GPT-6 Astra successfully completed Neal Agarwal’s ‘I AM Not a Robot’ game. According to Shameem’s post, the AI model progressed through the entire challenge and reached the final screen, where it received a “Human Verification” result.
The demonstration does not mean that AI can defeat every Captcha used by websites. However, it shows that modern AI models are becoming capable of handling interactive computer tasks rather than simply answering basic questions.
How Astra Handled The Captcha Game
Captchas are commonly used by websites to separate human users from automated software. They can ask people to identify objects in images, solve puzzles or perform actions that are intended to be difficult for bots.
Google’s reCAPTCHA is one of the most widely known examples, although several other companies also provide similar anti-bot technologies.
The game created by Agarwal uses a series of different challenges. Players have to identify visual elements, move objects, follow instructions and understand what each level is asking them to do. Completing the full game can take a human player around 30 to 60 minutes.
GPT-6 Astra reportedly used visual understanding together with its Computer Use capabilities to work through the challenges. It could examine the browser interface, understand the objective and perform actions such as clicking, dragging and typing.
As per the Shameem’s post, The test was not limited to simple image recognition. Some levels required the model to understand the rules of a particular puzzle and determine the correct way to complete it.
After completing the final challenge, the game displayed its “Human Verification” result and generated a Verified Human certificate.
Interestingly, OpenAI Head of Applied Research Boris Power also commented on Shameem’s post, saying, “True AGI can go everything a human can do on the computer.” The test therefore offers another example of the progress being made in AI computer-use capabilities.
-Times now
