Alternatives & comparisons
Mercer | Mettl Alternative for the AI Cheating Era (2026)
Mettl proctors hiring and skills assessments at volume, with AI monitoring and browser lockdown built for thousands of concurrent candidates. It does visible behavior well. But Cluely isn't visible to it, at any scale. Here is what sits below the observation layer.
Other exam-security alternatives to consider
No single tool covers every layer. Here is how the rest of the lockdown-browser and proctoring-service field compares on the same AI-cheating threat model.
What Mercer | Mettl Does
Mettl, part of Mercer, is a talent-assessment platform used heavily for corporate hiring, campus recruitment, and skills certification: assessment contexts that run at genuine volume, often thousands of candidates in a single hiring window rather than one exam hall's worth of test-takers. Its AI proctoring layer watches for face position and multiple-face detection, background noise, and tab-switch or copy-paste activity, paired with a browser lockdown for the assessment window. It works well for the things that layer can observe: an unauthorized person in frame, leaving the assessment tab, background conversation.
In short: Mettl is built to run webcam and browser-level checks at hiring scale, not to see what else is running on each candidate's machine.
What Mercer | Mettl Cannot See
Face detection, noise detection, and browser lockdown all operate on what a webcam and a browser tab can observe. An invisible AI overlay like Cluely marks itself excluded from the screen-capture APIs those checks depend on. A candidate can pass every face and noise check while reading answers off a surface Mettl's monitoring was never designed to capture. And because Mettl assesses at hiring volume, the same $20/month tool scales across every candidate in the pipeline at no additional cost to the person cheating.
The same limitation applies to on-device LLMs. A model running in local memory produces no webcam-visible behavior, no tab-switch event, and no network traffic. Aiseptor detects the technique instead of the appearance: a screen-capture-exclusion flag, GPU VRAM deltas, and DNS/SNI to AI endpoints, regardless of process name.
Scale changes the size of the risk, not its shape. The gap is the same one webcam-and-browser proctoring has everywhere else.
Comparison: Mercer | Mettl vs. Network-Layer Enforcement
| Capability | Mercer | Mettl | Aiseptor (Network Layer) |
|---|---|---|
| Face / noise / tab-switch detection | Yes | No |
| Browser lockdown | Yes | Complementary |
| Invisible AI overlays (Cluely, Pluely, unknown forks) | No | Yes |
| On-device LLMs (Ollama, LM Studio) | No | Yes |
| OS-wide network / DNS enforcement | No | Yes |
| Data collected | Webcam, audio, browser activity | Network access signals only |
Complementary Architecture
Mettl = webcam, environment, and browser-lockdown layer at hiring volume. Aiseptor = network and device layer. Mettl's face and noise detection catch environmental cheating Aiseptor does not address: an unauthorized person in frame, background conversation. Aiseptor catches the device and network AI surface a webcam cannot see, at the same hiring-pipeline volume Mettl already operates at. We don't claim to replace identity verification or physical proctoring.
See also: AutoProctor alternatives · Talview alternatives
Frequently Asked Questions
Does Mercer | Mettl detect Cluely?
Mettl's AI proctoring watches for face position, multiple faces, background noise, and tab-switching inside a browser lockdown. Cluely is an OS-layer overlay that excludes itself from the same screen-capture APIs that feed those checks. None of Mettl's monitoring signals (face detection, noise detection, tab focus) change when an overlay is rendering answers above the candidate's editor.
Does Mettl detect AI cheating generally?
Mettl's automated proctoring can flag behaviors visible to its webcam and browser layer: a second face, leaving the exam tab, unusual background audio. It cannot see an OS-layer overlay that hides from screen capture, or an on-device language model running in local memory with no network traffic.
Mettl assesses hiring candidates at high volume. Does that change the AI-cheating risk?
It raises it. Mettl is built for high-volume corporate hiring and campus recruitment, often thousands of concurrent candidates per assessment window. AI cheating tools scale the same way legitimate candidates do: a subscription overlay costs the same $20/month whether one candidate uses it or ten thousand do, and volume alone doesn't create a detection advantage for behavioral or webcam-based signals.
Is there a Mettl alternative without webcam?
Yes. Aiseptor enforces exam integrity at the network and device layer: no webcam, no face detection, no keystrokes. The audit record is metadata about what was and wasn't reachable during the session, not a recording of the candidate. Full data model on the trust page.
Where Aiseptor Fits: Beneath, Not Instead Of
Aiseptor is a layer, not a rip-and-replace. It sits beneath Mercer | Mettl, and beneath any lockdown browser or proctoring service, owning the device and network layer those tools architecturally cannot reach. Keep what you have for browser control or webcam proctoring; add Aiseptor for the OS- and network-level AI threats it was built to stop. The exam page can open in a normal browser while Aiseptor enforces the machine boundary.
What Aiseptor does not do: physical and environment security is out of scope. Aiseptor does not verify identity, watch the room, or catch a phone, a paper note, or an in-person accomplice off-camera. For those, pair Aiseptor with a live proctor or an identity-verification step. Aiseptor secures the device and the network path, not the physical room around it.
Volume changes the numbers. Not the blind spot.
Aiseptor pairs with Mettl's webcam and browser layer to close the network and device gap underneath, at the same hiring-pipeline scale. The audit record is metadata about what was and wasn't reachable, not a recording of your candidates.