A Redditor has demonstrated the ability of the recently launched ChatGPT agent to casually swat away a CAPTCHA, so that it can complete its assigned task(s). A couple of illustrative screenshots show the ChatGPT Agent handle, and commentate upon, dismissing Cloudflare’s ‘I am not a robot’ button (h/t Germany’s WinFuture).
Logkn’s Reddit post doesn’t provide any background details about the agent’s task at hand. However, we can see it was filling out some kind of web-based form, where it had to choose a translation target language. Before being able to click ‘convert’, it looks like the agent had to first pass Cloudflare’s bot protection.
“The link is inserted, so now I’ll click the ‘Verify you are human’ checkbox to complete verification on Cloudflare,” narrated the ChatGPT agent. “This step is necessary to prove I’m not a bot and proceed with the action,” it added without a hint of irony.
After sidestepping Cloudflare’s digital bouncer, the agent continued to dryly recount its progress. “The Cloudflare challenge was successful,” it informed the Redditor. “I’ll click the Convert button to proceed with the next step in the process.”
Cloudflare’s Turnstile system, presenting the simple ‘Verify you are human’ checkbox to end users, is one of the lower-friction annoyances for folks using the internet to get things done. However, its minimal nature and lightweight bot filtering stance are definitely not bulletproof as far as the new ChatGPT agent is concerned. WinFuture notes that Cloudflare’s system “checks user behavior such as mouse movements, click times, and browser fingerprints to distinguish between humans and machines” (machine translation). It might be a good idea for Cloudflare to apply a little tuning in light of this news.
What is the ChatGPT agent?
In case you missed it, OpenAI publicly introduced the ChatGPT agent in mid-July. The primary attraction of this agent is that, within the confines of “its own virtual computer,” the agent can be set and complete complex tasks from start to finish, says OpenAI.
The pioneering AI firm insists that the ChatGPT agent will always leave its user in control, asking permission “before taking actions of consequence,” and being easy to interrupt or stop in its tracks. You can read the introductory blog post or just watch the video embedded below for a summary of the ChatGPT agent’s abilities.
Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.
