Ofradr AI Review: Is It Detected on SEB & Mercer Mettl?

Ofradr has gained popularity as an AI helper for online assessments. But students frequently ask: is Ofradr actually undetectable on Safe Exam Browser (SEB) and Mercer Mettl, and why does downloading it trigger security warnings? Here is an objective technical breakdown.

What is Ofradr AI and how does it work?

Ofradr is an ~800KB native Windows application designed to function as an AI screen overlay. It captures screen content and provides AI-generated assistance for coding tasks and multiple-choice questions.

Ofradr attempts to render its overlay directly over assessment windows and offers options for both cloud-connected AI and local open-source models (such as Llama and Mistral).

Why Ofradr gets flagged on SEB and Mercer Mettl

While Ofradr aims to hide its overlay from basic screen sharing (like Zoom or Teams), strict proctoring platforms like Safe Exam Browser (SEB) and Mercer Mettl operate on a deeper level:

• Process signature detection: SEB maintains a process blocklist and monitors window hooks. Ofradr's binary signature is registered in SEB's blocked process database, causing SEB to refuse to launch or terminate the test.
• Activity logging & focus loss: Mercer Mettl logs window violations and focus switches. Interacting with Ofradr without a true zero-focus overlay triggers focus-loss count flags on Mettl's candidate report.
• Lack of hardware keystroke injection: Ofradr lacks true OS-level virtual hardware keystroke emulation. Inserting answers into locked code editors can trigger paste detection flags.

Antivirus and SmartScreen security warnings

Another common concern among candidates is that downloading Ofradr from ofradr.com frequently triggers Windows SmartScreen warnings, browser blocking, or domain reputation flags on security platforms like Gridinsoft. This is typical for unverified executables that intercept system inputs.

Local AI models vs. Frontier AI models

Ofradr promotes running local AI models (Llama/Mistral) offline. However, running local LLMs requires significant GPU and RAM resources, which can cause severe lag during timed exams. Furthermore, smaller 7B/8B local models often hallucinate or fail on complex algorithmic coding tasks compared to frontier web models like ChatGPT (GPT-4o) or Claude 3.5 Sonnet.

The stealth alternative: UFOlee

For candidates requiring 100% stealth on proctored exams, UFOlee provides an ultra-lightweight ~450KB Win32 architecture with zero taskbar icon or process signature match, hardware-level virtual keystroke typing, and direct connection to your official ChatGPT/Claude web accounts.

FAQ