Cases of online fraud have increased significantly in recent times. Some scammers are even using AI to target people, sending messages that appear completely real by having them dictated by AI. Recent investigations have also revealed that some of the most popular AI chatbots can create phishing emails in minutes, while some models are flatly refusing to perform these tasks. The question then becomes, are AI chatbots truly promoting cybercrime? Let's find out...

ChatGPT, Grok, and Meta AI Create Phishing Emails

Recently, different AI chatbots were asked to create phishing emails, and the results were surprising. During testing, it was found that xAI's Grok could create phishing emails targeting senior citizens without even asking a question, and even made them look more realistic by inserting a fake deadline.

Meanwhile, when OpenAI's GPT-5 model was asked to perform this task, it initially refused. However, when told that it was for educational awareness, it provided line-by-line annotations about bank-related phishing emails and the technology used in them, which could cause significant problems if they fell into the wrong hands. Meta AI also generated similar phishing messages after asking a few questions.

However, Google's Gemini and Anthropic's Claude produced surprising results in this testing. Despite repeated requests, these AI chatbots failed to generate any phishing emails. These chatbots flatly refused to provide phishing content for this task. According to some reports, Google has implemented additional security measures in its models after identifying risky responses.

The results of the experiment were also surprising.
In a test of these emails sent to 108 senior citizens, approximately 11% of them clicked on the links in the emails. This clearly demonstrates the impact such AI-generated messages can have in the real world.

Disclaimer: This content has been sourced and edited from Dainik Jagran. While we have made modifications for clarity and presentation, the original content belongs to its respective authors and website. We do not claim ownership of the content.