Attackers Automate API Key Theft Using 1,438 Fake FOFA Accounts
A threat actor has created a sophisticated automated system using 1,438 fake FOFA accounts to systematically steal exposed OpenAI and DeepSeek API keys across the internet.
A novel and automated threat campaign is systematically harvesting exposed API keys for OpenAI and DeepSeek by leveraging a large network of fake accounts on the FOFA search engine. Researchers at VECERT have identified this operation, dubbed "Operation Ghost-Campus," which utilizes 1,438 distinct FOFA accounts to bypass the search engine's query limits and scan the internet for exposed secrets. The stolen API keys are then used to power the attacker's own AI tools and scripts, effectively providing them with free computing resources at the expense of unsuspecting victims.
The core of this attack lies in the attacker's ability to automate the creation and use of FOFA accounts. By operating a private "catch-all" email system on the domain melocc.cc, the threat actor can rapidly register new FOFA accounts, receive confirmation emails, and activate them within seconds. This allows for a continuous stream of queries, enabling the scanning of misconfigured servers, exposed .env files, Docker settings, and poorly secured Git repositories for sensitive information.
Once potential API keys are identified using regular expressions, the system automatically tests them against the respective AI service endpoints. For OpenAI keys, it looks for those starting with sk-proj-, while DeepSeek keys are matched against the pattern sk-[a-f0-9]{32}. The system categorizes keys based on their validity and remaining credit, sorting them into working keys, those with depleted balances, and revoked credentials. This automated validation process ensures that only functional keys are collected for exploitation.
The stolen API keys are not merely hoarded; they are actively used by the attacker. VECERT analysis revealed that the keys power an interactive web terminal named "Pi — AI Coding Assistant," built using GoTTY, and support other scripts hosted on the attacker's server. This means victims not only face the risk of unexpected costs due to drained credits but also that their own API keys will appear in logs, potentially obscuring the true source of the abuse.
Beyond API key theft, the same server infrastructure used in this campaign also contained tools specifically targeting the Shanghai Publishing and Printing College. VECERT reported that data belonging to 5,132 students, including personal identifiers and campus card balances, was collected. Furthermore, bots were observed faking activity to accumulate school points, highlighting a multi-faceted malicious operation.
This campaign underscores a growing trend where attackers are scaling up traditional credential theft techniques by integrating them with AI services and automated infrastructure. The use of public search engines, disposable email accounts, and basic pattern matching demonstrates how relatively simple methods can yield significant results in the current landscape of exposed cloud credentials.
To mitigate such threats, organizations are advised to rigorously protect API keys by avoiding their inclusion in .env files, public repositories, or web-accessible folders. Regularly revoking and rotating any potentially exposed keys, coupled with vigilant monitoring of usage logs for anomalous activity, is crucial. Implementing spending limits on AI accounts and staying informed about the infrastructure associated with these campaigns can further enhance defenses.