How Modern Bots Fake AI Agent Traffic And How to Detect It 

by ·

Modern AI agents use specific tools to break down high-level tasks, such as website browsing or product comparison, and complete them independently.

A prominent security risk is that malicious bots make traffic look like it comes from a legitimate AI agent, potentially inheriting the trust from ChatGPT, Perplexity, or Meta. That can create security risks such as fraud, phishing, prompt injection, data leaks, etc.

DataDome found 16.4 million spoofed requests claiming to be Meta-ExternalAgent and 7.9 million claiming to be ChatGPT-User in early 2026. And almost 2.4% of traffic claiming to be PerplexityBot was also fraudulent.

Below, we’ll look at how bots are doing it and how fake agent traffic can be detected to stop AI scrapers.

How Modern Bots Fake AI Agent Traffic

The easiest method is identity spoofing. Many AI agents identify themselves using information contained in an HTTP request, including:

  • User-Agent strings
  • Referer headers
  • IP address
  • Browser information

These can be declarations, so they can’t be taken as proof. 

Bots can change their User-Agent to one of a known AI crawler or add a Referer that makes traffic appear to come from an AI service. And they know how to make life more difficult. The once-reliable standard IP checks don’t work for malicious bots because they can rotate through residential proxies and large IP pools to avoid basic reputation rules or per-IP rate limits.

Modern automation can also execute JavaScript, maintain sessions, and imitate normal browser behavior, so the depth of the issue only keeps on growing.

Detecting Fake AI Agent Traffic From Bots

One signal isn’t reliable enough to separate every genuine AI agent from every fake one.

Using Agentic Bot Detection and Management Services

Agentic AI bot detection and management services look at multiple signals instead of automatically trusting a recognizable agent name.

These can include:

Trust can then be reassessed as a session develops rather than decided from the first request.

Using Identity Signals

Identity signals can help establish whether an agent is actually who it claims to be. 

User-Agent strings are weak because anyone can copy them. Stronger signals include:

  • Official IP ranges (although not as reliable now)
  • Reverse-DNS validation
  • Cryptographic verification
  • Delegation and actor binding
  • Continuous just-in-time (JIT) attributes

Note: stronger proof of identity doesn’t prove the agent’s current actions are safe.

Network and IP Reputation

Security systems can analyze whether traffic comes from:

  • Expected IP ranges
  • Hosting providers
  • Residential proxy networks
  • Suspicious ASNs
  • IPs previously associated with abuse

So it’s deeper than just looking at individual IP addresses from traffic. They can also look for inconsistencies. An agent claiming a trusted identity while continually switching between unrelated residential and hosting networks could warrant closer inspection.

That said, IP reputation still shouldn’t make the final decision.

Behavioral Biometrics

Behavior can expose automation even if identity and network signals look legitimate.

Behavioral analysis can examine mouse movements and clicks, touch interaction, keystrokes and scrolling, request frequency, navigation sequences, and repeated access to sensitive endpoints.

A bot might imitate one action, but reproducing realistic behavior consistently across an entire session is harder. Think of how easily LLMs slip-up when you give them too many instructions in one go. 

Behavioral checks also help establish intent. A verified agent browsing normally before suddenly hammering checkout or account endpoints shouldn’t remain trusted simply because its identity hasn’t changed.

What to Do If You Suspect Fake AI Agentic Traffic?

Never blindly trust a known AI user agent or referer. Instead, use multiple signals to verify identity: signatures, published IP ranges or reverse DNS.

Don’t check just IP, ASN, proxy and session reputation but check them all and monitor request frequency and user behavior throughout the session.

Instead of permanent allowlists, impose rate limits on suspicious traffic and restrict access to sensitive endpoints. Regularly review trusted agents.

Using bot detection and management can also prevent bots before you notice them.

Modern bots faking agentic AI traffic present a severe structural crisis for businesses. Studies show AI crawlers and bots now make up more than 1 in 10 verified bot requests (DataDome), with attackers essentially weaponizing next-generation Large Language Models (LLMs) to fake human and agentic behaviors.