Recently, a new type of phishing attack targeting mainstream artificial intelligence tools has been occurring frequently. Hackers have targeted high-traffic AI services such as ChatGPT and Claude, using their official content sharing and plugin mechanisms to place indistinguishable phishing links in the top search engine advertisements. Since the root domain of these malicious landing pages is completely identical to the official one, ordinary users find it extremely difficult to identify them, leading to a significant increase in the success rate of clicking and downloading malicious installation packages.

In this exposed phishing method, hackers cleverly utilized ChatGPT Canvas's canvas creation and content sharing features to create and share a highly realistic official website page under the official domain. After users click into it, a pop-up window claims, "The current access volume is too high, please download the desktop version to continue accessing." However, this诱导下载 button actually points to a pre-prepared malicious Trojan program.

This deceptive method has repeatedly succeeded because it perfectly exploits users' trust. First, the hackers generate and share the content through ChatGPT.com's own canvas function, so the generated web link carries an unassailable official domain. Second, the top ads placed by hackers on Google search show a clean official domain URL, which makes users lower their guard before clicking. Finally, even if users are guided to the phishing page, the browser's address bar still displays ChatGPT.com without change, making it difficult for mainstream security software to directly intercept it as a malicious site.

Similarly, this kind of attack path has also been found on the Claude platform. Hackers use Claude.Ai's conversation sharing mechanism to pre-produce malicious conversation records targeting various popular computer software (such as hardware detection tools like CPU-Z). When netizens search for these software, the top fake advertisements will direct them to the official sharing chain of Claude.Ai, and then further induce listeners to download malicious software through fraudulent text in the conversation.

Regarding this kind of subtle attack method that relies on the "endorsement" of large companies, industry experts admit that due to the difficulty for platform providers to conduct pre-review of massive user-generated and shared content, such phishing activities are difficult to be completely cut off in the short term, and may even show a significant growth trend in the future. For ordinary internet users, in addition to being vigilant and refusing to download unknown "desktop clients," installing reliable ad-blocking extensions in the browser and directly blocking top search engine promotional ads at the source may be the most effective self-protection measures against such high-tech fraud currently.