StealthGPT Unveils Super Model to Evade Pangram Detection

Pangram has consolidated its position as the de facto authority on AI authorship, a status solidified by its integration into platforms like Substack. This technological gatekeeping has real-world consequences; a recent high-profile case saw a book contract terminated after Pangram flagged a manuscript as 78 percent AI-generated. StealthGPT founder Jozef Gherman frames the release of Super as a necessary response to an escalating arms race, arguing that individuals should retain control over how they leverage AI in their creative workflows.

Developed by lead AI engineer Danil Shashkov, Super shifts the technical approach from simple word substitution to the reconstruction of underlying neural patterns. Instead of merely purging common AI indicators like the word "delve," the model forces the text through a training architecture designed to mimic human stylistic variability. Internal testing conducted in September 2026 supports this, with the company reporting that Pangram v4 classified 89 percent of Super's outputs as human-written. This performance significantly outpaces other market competitors, which StealthGPT testing placed at 32 percent or lower in similar benchmarks. Detailed methodology and scoring data have been made available via the company’s public repository.