Recently, hackers are actively using popular artificial intelligence tools such as ChatGPT and Claude for phishing. Hackers choose these tools because they have high search volume and can reuse official domain names through plug-ins or content sharing. At least for most users, they cannot tell that this is a phishing website directly from the top advertisement of Google search results, so the success rate of hackers' phishing is significantly higher.
Malicious use of plug-in functions for phishing:
The ChatGPT Canvas function can be used to create a canvas similar to a web page. In the latest phishing campaign, hackers used this function to create a fake ChatGPT official website. In the canvas, the hacker claimed that the current traffic volume was too high and asked users to download the desktop version to continue visiting. The download button for the desktop version actually pointed to a malicious installation package provided by the hacker.
This phishing method has several notable features: 1. Hackers use the ChatGPT canvas to create content and share it, so the content link is still ChatGPT.com; 2. Hackers place ads in Google search, and the address displayed in the ads is also ChatGPT.com, which can reduce users' vigilance; 3. After users click on the ads to enter the landing page, the address bar still displays ChatGPT.com, and this website will not be blocked by security software.
It is also unlikely that OpenAI will review the sharing of content created by users, so such phishing attacks should increase significantly in the future. It may be a good choice for users to directly install ad blocking extensions to block ads from Google search or other search engines.
There are also similar fishing cases in Claude:
Some users have noticed that phishing websites related to Claude also appear in Google search ads. The phishing method is similar to the ChatGPT case, and also uses the content sharing function of Claude.Ai. Hackers first use Claude to create malicious conversations, which contain links to phishing websites or malware downloads controlled by hackers. Hackers create content sharing for a variety of popular software in advance. For example, when users search for CPU-Z, the top advertisement on the homepage will guide users to jump to Claude.Ai share content and then induce users to click on the link to download malware.
Anthropic obviously does not review the content shared by users, so this phishing method is actually difficult to completely cut off, which is why directly blocking various search engine ads may be the best approach.
via Jan