The Aiseptor agent collects the following data only for the duration of an active exam session. Data collection begins when the candidate joins (VPN tunnel established) and ends when the session closes (tunnel torn down). No persistent agent remains on the device after session close.
4.1 Network-layer telemetry
- Network access request records and timestamps (for policy enforcement; no payload contents are captured)
- Session cryptographic identifier (derived per session; rotated each exam)
- Assigned ephemeral VPN IP address
- Outbound connection destinations that fall outside the exam whitelist
4.2 Device security signals
- Running process names (checked against a known-AI-tool denylist; full process list is not transmitted)
- GPU VRAM utilization delta (used to detect on-device LLM inference; raw readings are not stored)
- Presence of known AI model file extensions (.gguf, .safetensors) at standard install paths
- Access policy and network configuration integrity status
4.3 What we do NOT collect
- Webcam or microphone streams
- Screen recordings or screenshots
- Keystrokes or clipboard content
- Browser history outside the exam session
- Personal files, documents, or application data
4.4 Purpose and legal basis
Candidate exam data is collected solely to generate a per-session integrity report (the “Aegis report”) delivered to the administering institution. The legal basis for processing is legitimate interest: specifically, the integrity of the assessment that the candidate voluntarily enrolled in; and, where applicable, performance of a contract between Aiseptor and the platform customer.
Candidates are informed of Aiseptor's involvement by the administering institution before the session begins. If you were not informed, contact the institution that administered your exam.
4.5 AI signal processing
Device security signals are analyzed in real time by automated classifiers to produce integrity flags (e.g., AI_LOCAL_LLM, GPU_VRAM_ANOMALY), the same class of on-device AI detection signals documented on our AI-overlay detection page. No individual signals are shared with external AI model providers. Classification occurs on Aiseptor-controlled infrastructure.