Cheating in 2025: AI and Non-AI Methods

AI-enabled cheating methods

• Digital deepfakes — Real-time face/voice manipulation to defeat ID verification and liveness checks during onboarding or re-authentication.
• Hidden cameras feeding AI — Mini cameras (e.g., in glasses or buttonholes) capture items; an AI tool parses the question and returns answers via earpiece or smart glasses.
• Generative AI apps, plugins, and extensions — Browser helpers read what’s on screen and serve answers in real time.
• Overlay software — On-screen assistants that read visible text and prompt responses, often hard for screen-recording/mirroring tools and some secure browsers to detect.
• AI voice assistants — Candidates whisper a question and receive answers as discreet audio prompts.
• Agentic browsers — AI-augmented browsers summarise stems and propose answers directly in the window.
• Computer-use models / autonomous agents — Agents take control of the browser to complete interactive tasks or assignments on the candidate’s behalf.
• Screen-share hardware + AI — Inline HDMI splitters or capture devices mirror the exam to an external system, where AI reads items and relays answers via audio or vibration.
• DWM hijacking (AI-assisted analysis) — Hooking Windows Desktop Window Manager to capture the screen stream before secure-browser controls bite, then using AI to parse content quickly.
• Bypassing content protection (AI-assisted extraction) — Man-in-the-Middle tactics or Accessibility APIs to exfiltrate test content at scale for AI processing.
• Virtual machines (paired with AI) — Hardened VMs evade secure-browser detection, allowing forbidden apps (including AI tools) to run concurrently.

Non-AI cheating methods

• Imposter/proxy test taking — Another person sits the exam for the candidate (in person).
• Virtual proxy test taking — The candidate remains on camera, but a remote helper controls the desktop and inputs answers.
• Simple impersonation — A look-alike (e.g., a relative) uses the candidate’s ID.
• Covert audio communications — Magnetic/capsule earpieces play pre-recorded info or relay a helper’s guidance.
• Concealed notes — Crib sheets hidden in permitted items (pencil cases, modified “super pens” with LCDs, rolling-paper pens).
• Smart devices (non-AI use) — Cheating watches, smartwatches, or advanced calculators preloaded with notes or used for basic communications.
• Smart rings — Bluetooth rings enabling two-way audio.
• AR contact lenses (non-AI use) — Discreetly display stored prompts or notes.
• Concealed AirPods — Answers fed verbally, hidden by hair or clothing.
• Bone-conduction headsets — Worn discreetly (even strapped to a limb); answers conveyed as audio cues or Morse-style vibrations.
• Wireless messaging via “magic” calculators or custom watches — Send/receive messages or emails during the test.
• Visual recording/photography — Hidden cameras in clothing, pens, or glasses to harvest items for real-time collusion or later reuse.
• DWM hijacking / content capture (non-AI use) — Using hooks purely to record/stream content without downstream AI.
• Bypassing content protection (non-AI use) — Network or OS-level exploits used purely for extraction.
• Virtual machines (non-AI use) — Run forbidden apps/resources undetected alongside the secure browser.

Reality check: Many “non-AI” methods are easily supercharged by AI. The lists above reflect the primary mode of each tactic, not its possible combinations.