Identifying Agentic Automation With Behavioral Telemetry
Akamai Technologies, Wednesday, August 19th, 2026
Akamai research shows AI browser agents evade classic bot detection, and tests a transformer model that spots them anyway.
Akamai security research examines a new class of web traffic generated by autonomous AI browser agents such as Perplexity Comet, OpenAI Atlas and the Claude Chrome extension.
These agents produce extremely sparse behavioral telemetry, with more than 98% of their requests containing little or no mouse movement, which defeats traditional bot detection models. Akamai built a proof-of-concept Masked Autoencoder Transformer that treats mouse telemetry like language tokens and analyzes short event sequences.
A lightweight classifier trained on the resulting embeddings separated agentic automation from sparse human sessions. The model achieved high accuracy despite extreme class imbalance in the data.