"AI Content... Like Humans" Detection Evasion Services on the Rise

AI Detection Evasion Service $9~57 per Month
Even Big Tech Can't Distinguish Real from Fake

Services that disguise AI-generated content as human-made are increasing.

According to the information technology (IT) industry on the 7th, US AI company WordAI is supplying AI detection evasion services for $9 to $57 per month.

OpenAI logo. Photo by Reuters [Image source=Reuters Yonhap News]

The company promoted that AI-generated content processed through its service is not detected by AI detection technologies such as OpenAI, Originality.ai, and Hugging Face. AISSEO and Frazzly.ai are also operating similar evasion services.

Services that help evade AI detection are also called 'AI humanizers,' meaning they make the style and expressions naturally appear as if written by a human. The main customers of these services are companies trying to advertise using spam emails and Google searches.

On the internet, methods to disguise AI-written texts as human-written can be easily found. Advice ranges from varying sentence lengths and mixing in everyday vocabulary to deliberately including grammatically incorrect expressions to reveal mistakes a human might make.

Big tech companies are struggling to filter AI content. On the 1st, Meta decided to replace the 'made with AI' label on social media platforms such as Instagram, Facebook, and Threads with 'AI info,' implying that AI assistance was suspected in some parts of the production process. This change came after photographers found AI-made tags attached to photos they had actually taken. The industry suspects that applying Adobe’s photo editing tools causes these AI label errors.

There is also analysis that determining AI content has become practically difficult. A research team from the University of Reading in the UK published last month in the scientific journal PLOS ONE the results of a study where they submitted 33 sets of answers generated by ChatGPT-4 for five short-answer and essay exams required for a bachelor's degree in psychology.

The study found that 94% of the AI-generated answers were not detected by graders, and there was an 83% probability that AI answers scored higher compared to actual student responses.

This content was produced with the assistance of AI translation services.