🤖 “I am not a robot,” ChatGPT Agent checks, then proceeds to bypass a Captcha. A reddit user on Friday shared the absurd moment they witnessed OpenAI’s ChatGPT Agent pass the very security measures designed to stop it in its tracks. The pop-up we’ve all come to know well before clicking our way through most websites proved ineffective in the face of AI, putting into question the efficacy of bot-detection systems on the internet, according to ARS Technica.

Acting on behalf of its users, the ChatGPT Agent can perform multistep tasks and access the internet within a secure environment. The new OpenAI feature allows users to watch as the agent navigates the browser through a window in the ChatGPT interface. The user reporting the incident watched as the agent clicked on the “verify you are human” checkbox, while ironically stating that “this step is necessary to prove I’m not a bot” as part of its real-time action narration protocol.

The Cloudflare checkbox for anti-bot verification doesn’t just rely on a simple click, it employs a complex behavioural screening system that analyzes mouse movements, click timing, browser fingerprints, and IP reputation to test for correspondence to human behavior. What’s astonishing is that an actual bot was able to evade a Captcha altogether — an escalated security measure specifically made to tell bots and humans apart through obscure visuals indecipherable for a machine — which users are often met with when suspicious activity is detected.

AI vs Captcha is not new, however. Captchas have been around for quite some time, and have had their share of defeats to AI. More irony lies in the fact that ever since the reCaptcha project was acquired by Google in 2009, it helped train AI models for image recognition, ultimately contributing to AI now being able to bypass Captchas.

Are Captchas about to become a thing of the past? The incident gives us a glimpse into our future with AI, suggesting that its capabilities have surpassed simple scripted automation and are brushing up against human judgement. If AI can masquerade as human, how long will it be until bot-detection becomes obsolete? That, for now, remains uncertain.

Tags: