OpenAI’s new ChatGPT agent, which can perform multi-step tasks for users in a browser, has proven it can pass one of the most common checks on the Internet. It passed Cloudflare’s CAPTCHA, a checkbox that is supposed to block bots from accessing content.
OpenAI’s new ChatGPT agent, which can perform multi-step tasks for users in a browser, has proven it can pass one of the most common checks on the Internet. It passed Cloudflare’s CAPTCHA, a checkbox that is supposed to block bots from accessing content.
ChatGPT Agent allows OpenAI’s AI to control a web browser, running in an isolated environment with its own virtual operating system and browser that has access to the real Internet. Users can observe the AI’s actions through a window in the ChatGPT interface, controlling the agent’s work as it performs tasks, notes Ars Technica.
The system requires user permission before performing actions that have real consequences, such as making a purchase. Recently, a Reddit user with the nickname «logkn» demonstrated that they discovered that ChatGPT Agent is capable of something ironic.
He showed screenshots of the AI agent easily passing the CAPTCHA (abbreviation for Completely Automated Public Turing tests to tell Computers and Humans Apart) verification stage while performing video conversion tasks.
The agent went through a two-step verification process: first, he clicked the «Confirm you are human» checkbox, and then, after successfully passing Cloudflare’s verification, he clicked the «Convert» button.
«The link has been inserted, so now I will click the „Confirm that you are human“ checkbox to complete the verification on Cloudflare. This step is necessary to prove that I am not a bot and continue,» the AI commented on its actions.
Although the agent in this case did not encounter a real CAPTCHA image puzzle, it successfully passed Cloudflare’s behavioral monitoring, which determines whether to apply more complex validation when it detects automated activity.
The system, called Turnstile, analyzes several signals: mouse movements, click times, browser fingerprints, IP address reputation, and JavaScript execution patterns to determine whether a user is exhibiting human-like behavior. If these checks pass, users continue without seeing the CAPTCHA puzzle. If the system detects suspicious patterns, it moves on to visual puzzles.
The ability of AI to bypass some CAPTCHA checks is a concern for developers of this protection system, who are working to improve it, but bots have been able to bypass this check before. CAPTCHA is now seen more as a way to slow down bot attacks or make them more expensive, rather than as an impenetrable defense.
As a reminder, OpenAI recently launched a new AI agent in ChatGPT that can perform a wide range of tasks on behalf of users. The company says the agent can automatically navigate a calendar, create editable presentations and slideshows, and run code.
