Mechanisms that use symbols invisible to the human eye in text have previously been used to transmit hidden malicious requests to artificial intelligence systems. Now, pay attention Microsoftwhich is used to send spam, which is not easy for modern artificial intelligence services to combat.
Image source: microsoft.com
A set of 128 Unicode characters in a special block that is invisible to humans but can be read by computers. Previously, these characters could be embedded in emails: people viewed them as harmless text, and models analyzing these letters were given malicious instructions. Microsoft says that now the mechanism has gone in the opposite direction – symbols invisible to humans help fool spam filters. Sometime in early February, Microsoft Defender for Office noticed that the number of emails with hidden characters jumped from 21,000 to 1.3 million emails per day. Within four days, the number increased to 2.5 million, and the flow continued for several months before declining sharply at the end of May.
Spam filters often search letter text for keywords: “credit,” “term,” or instructions for a specific amount – all of which are often found in mass mailings. To trick them, spammers began inserting symbols invisible to humans between words. As a result, the text of the letter remains the same to a person, but the computer no longer sees the keywords, even though these words are actually used in the text of the letter. Artificial intelligence models connected to spam filters based on machine learning and natural language processing mechanisms will not respond to them either. The only obvious way to combat this scourge is optical character recognition tools, which open text the way people see it.
Microsoft warns that traditional spam filters may be powerless against such mechanisms and provides recommendations for more effectively detecting ASCII character substitutions in spam messages.
If you find an error, select it with your mouse and press CTRL+ENTER.