ChatGPT Dots repeatedly prompts for GitHub create_blob approvals even after Always allow. We inspect the ~720 KB size boundary, local computer access risks, and stopping controls.
A reverse-engineering report decrypted the obfuscated bytecode running on ChatGPT's login page and uncovered the 55-field fingerprinting system and React Fiber inspection behind Cloudflare Turnstile.
I investigated the source behind the viral claim that a Johns Hopkins study found ChatGPT lies 27% of the time, and it turns out multiple different studies have been mixed together.