AI Agents Mimic Humans, Undermining Traditional Bot Detection
What Happened — In a Help Net Security interview, Zohaib Ahmed, CEO of Resemble AI, explained that modern AI‑driven agents can operate in three modes—declared, undeclared, and disguised. Unlike classic scripted bots, these agents choose their own steps toward a goal (e.g., booking a flight) and can alter their browser fingerprints to appear human. Traditional binary bot detection (human vs. bot) often misses undeclared agents and almost always misses disguised agents. Ahmed highlighted behavioral signals that still surface: click timing, pointer movement, typing cadence, scrolling patterns, session‑level goal convergence, and OAuth consent logs. He warned that employees increasingly embed agents into email, calendars, and SaaS tools via personal API keys and OAuth grants, creating “shadow” traffic that security teams may not see.
Why It Matters for Trust & Control Assurance — The story tests the control objective of AI governance and monitoring, showing that without dedicated detection signals, organizations cannot reliably evidence that AI‑driven automation is under oversight.
Who Is Affected — SaaS providers, cloud‑based enterprises, and any organization that integrates third‑party APIs or automation into business workflows.
Recommended Actions
- Map AI‑agent behavioral signals (click timing, session swarms, OAuth consent logs) to your existing monitoring controls.
- Extend logging to capture OAuth grant usage and outbound API traffic for anomaly detection.
- Incorporate AI‑agent detection into your security awareness program to surface “shadow” automation.
Technical Notes – Detection signals include pointer‑move entropy, click‑sequence latency, typing flow, scroll velocity, request‑header anomalies, and Web Bot Auth signatures. Disguised agents deliberately spoof user‑agent strings and browser fingerprints, rendering static fingerprinting ineffective. Source: https://www.helpnetsecurity.com/2026/11/05/zohaib-ahmed-resemble-ai-detecting-ai-agent-traffic/