Russian hackers are bypassing ChatGPT restrictions

OpenAI tried to build a digital moat around ChatGPT to keep it out of the hands of cybercriminals in restricted regions. But in underground hacking forums, that fence barely registered as a speed bump.

Security researchers at Check Point Software recently spotted Russian threat actors actively sharing bypasses for OpenAI's restrictions. Scouring cybercrime hubs, the team found discussions detailing how to defeat geo-fencing, IP filters, phone verification, and payment blocks.

Virtual numbers for two cents

To lock down the chatbot, OpenAI relies on geo-blocking that keeps Russian IPs out and requires valid mobile numbers for SMS activation. Underground forums solved that within weeks.

Russian-language tutorials on these boards walk users through semi-legal online SMS portals where a disposable foreign number costs just 2 cents. That is all it takes to receive an activation code and spin up an account. In other threads, actors discussed using stolen payment cards to fund API access, allowing them to bypass regular web interface friction entirely.

Sergey Shykevich, threat intelligence group manager at Check Point, noted that bypassing these controls simply is not difficult. To attackers, ChatGPT represents a drastic drop in operational costs.

Basic infostealers and prompt trickery

Once inside, these actors are not just playing with prompts. They are testing code.

In one forum thread titled "ChatGPT – Benefits of Malware," a member demonstrated how feeding published malware analysis into the bot yielded a working Python infostealer. The script crawled common file types, bundled them into a random folder, and dumped them onto a hardcoded FTP server.

Attackers are also leaning on the model to draft polished phishing lures that shed telltale grammatical errors. While OpenAI notes it moderates misuse, it acknowledged that some requests slip through the cracks, often disguised as harmless academic research or fictional creative writing.

Why it matters

This is a stark reminder that standard SaaS defenses like geo-fencing and SMS verification fold instantly against motivated attackers.

For everyday defenders, the risk is not that AI is inventing hyper-advanced super-malware overnight. The real danger is the automation of the baseline. Novice actors who previously lacked English fluency or scripting skills can now spin up functional stealer code and convincing phishing campaigns for pocket change.

OpenAI finds itself in an uphill battle. Surface-level geo-fencing and superficial keyword filters end up acting like digital theater. We believe model providers will quickly have to move past basic perimeter security and build far deeper intent analysis directly into inference pipelines, because attackers will always find a two-cent path around the front door.