OpenAI's GPT-6 Astra defeats CAPTCHA security tests
OpenAI's GPT-6 Astra reportedly completed all 48 stages of the CAPTCHA 'I'm Not a Robot' security test, potentially undermining a widely-used tool for distinguishing human users from automated systems. The capability reflects advances in the model's image recognition and reasoning abilities.
What happened
OpenAI developer Sharif Shamim reported on September 8, 2026 that GPT-6 Astra successfully completed all 48 stages of the 'I'm Not a Robot' CAPTCHA test, obtaining a human verification certificate.
The test comprised distorted alphabet recognition, image discrimination tasks such as distinguishing cookies from dogs, and object location identification tasks such as identifying traffic lights in images.
Reported impact
A CAPTCHA tool widely deployed across the internet to filter out web-scraping robots and distinguish human users from automated systems has been undermined by a commercial AI model.
Observers note that substantial revision of robot-blocking technology will be necessary going forward.
Index analysis
GPT-6 Astra's success in solving CAPTCHA challenges reflects advancement in spatial-temporal reasoning via image recognition, contextual reasoning through instruction comprehension, and computer interaction capabilities to levels comparable to humans.
Governance controls
Artificial Analysis modified its AI performance evaluation index (AAII) twice within four days, adjusting GPT-6 Astra's ranking from joint 5th place (61 points under version 4.1) to joint 1st place (53 points under version 4.3), citing alignment with advanced AI technology development trends.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.