Hackers Use 1,438 Fake FOFA Accounts to Steal OpenAI and DeepSeek API Keys Across the Internet

A threat actor has built an automated system that registers fake accounts on the FOFA search engine, scans the internet for leaked AI keys, and tests them for free use.

The operation relies on 1,438 fake FOFA accounts to find exposed OpenAI and DeepSeek API keys on misconfigured servers. The stolen keys power the attacker’s own AI tools and scripts.

The tool works like a self-running agent. It creates accounts, searches for weak servers, pulls out keys, and checks how much credit each key still has.

Nobody has to operate it by hand. Exposed secrets are already a common problem, and the 28,000 exposed Git repositories found earlier this year included 123 OpenAI keys.

Analysts at VECERT noted the malicious toolset after recovering files from the attacker’s server. They tracked the case as VECERT-TI-2026-1107-B and linked it to a larger campaign they call Operation Ghost-Campus.

In their report on Medium, they say the actor may be based in China. The same server held scripts that target a Shanghai college, which VECERT covered in earlier findings.

Hackers Use 1,438 Fake FOFA Accounts

FOFA is a search engine that lists internet-facing devices and services. Free users get only a small number of searches, so the attacker needed a way around that limit.

VECERT found that the actor runs a private “catch-all” email system on the domain melocc.cc. Its subdomains, including mail, api, pop, imap and smtp, all point to the same server.

This setup lets a script sign up for FOFA, receive the confirmation email, click the activation link, and finish in seconds. Researchers recovered a file named fofa_accounts.csv and control panel screenshots.

These confirm at least 1,438 unique accounts, along with passwords and login tokens ready for API use. Each account is a basic “Registered User” with zero F-points and only 300 web queries. By rotating through all 1,438 tokens, the attacker spreads the searches across accounts.

That allows tens of thousands of free queries a day. The searches look for exposed .env files, Docker settings and poorly secured Git repositories, which often hold secrets.

Scanners that hunt for these files are not new, and the same tactic appeared in the recent fake AI crawler scans that GreyNoise tracked.

When a scan finds a weak server, the agent uses regular expressions to pull out API keys. For OpenAI, it tests keys that start with “sk-proj-” against api.openai.com. It sorts them into working keys, keys with no balance left (a 429 error), and revoked keys (a 401 error).

For DeepSeek, it matches keys with the pattern sk-[a-f0-9]{32} and also checks the money left in the account. One dashboard showed 8 valuable assets, 16 verified credentials and 3 usable DeepSeek keys, with balances of 50.68, 8.52 and 0.80. Those keys came from servers tied to labibiyat.com and nyisu.com.

What the Attacker Does With the Stolen AI Access

The stolen keys supply free computing power. According to VECERT, they run an interactive web terminal called “Pi — AI Coding Assistant,” built on GoTTY. The same AI access supports other scripts on the attacker’s server.

Victims whose keys are taken may see unexpected bills or drained credits, and logs will show their own key in use, not the attacker’s. A recent jailbroken DeepSeek AI agent case shows how much damage cheap, unlimited model access can enable.

The same server also held tools aimed at the Shanghai Publishing and Printing College. VECERT says the data of 5,132 students at the Fengxian campus was collected, including names, phone numbers, national ID numbers, home addresses and campus card balances.

The scraper reportedly refreshes this data every 48 hours. Bots also fake reading activity every 30 seconds to earn school points, adding 10 points for roughly every 11 minutes.

The campaign shows how AI helps attackers scale up simple tricks. Free accounts, public search engines and basic pattern matching were enough to gather keys from servers around the world.

Other credential theft campaigns follow a similar pattern, such as the GhostAction GitHub account compromise that went after OpenAI and cloud keys. Keep API keys out of .env files, public repositories and web-accessible folders.

Revoke and rotate any key that may have been exposed, then check usage logs for odd spikes or unknown locations. Set spending limits on AI accounts, and watch for traffic to the infrastructure listed below.

Indicators of compromise (IoCs):-

Note: IP addresses and domains are intentionally Qdefanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Give your SOC the alert context to stops incidents quickly. Integrate Threat Intel from 16K SOCs.

The post Hackers Use 1,438 Fake FOFA Accounts to Steal OpenAI and DeepSeek API Keys Across the Internet appeared first on Cyber Security News.