Cequence says nearly all fake AI-bot requests in its protected traffic were blocked or failed

Cequence Security reported that one scanner impersonated about 32 AI-bot identities while probing roughly 90 paths for credentials and other sensitive files. The company said nearly all of the requests in traffic it protected were blocked or failed. Cequence Security’s report was published on October 28, 2026; Cequence observed the activity from at least late September through late October 2026.

How the fake AI-bot identities worked

The scanner cycled through identities associated with OpenAI, Anthropic, Meta, Perplexity and other providers. Cequence observed a single machine claiming identities associated with competing providers within seconds. Some fake Claude identities reused Anthropic’s crawler contact details, while the scanner offered no identifying information or contact address of its own.

A user-agent is a request field that tells a website what kind of software is making a request. Its text is self-declared, so the label alone does not verify who sent it. Cequence recommends checking an agent’s identity through signed requests, published address ranges or a direct integration, then assessing whether its behavior fits the identity it claims.

What the scanner sought, and what happened to requests

Cequence counted roughly 90 targeted paths, including .env configuration files, cloud credential files, private SSH keys and server environment data. Before requesting sensitive paths, the scanner tried random, meaningless URLs to find websites that returned a response for every request. Each scanning address then sent between 50 and 190 requests in about eight seconds before moving on.

In the traffic Cequence protected, nearly all of the scanner’s requests were blocked or failed. That is the reported outcome for that traffic. The activity was credential-seeking; the report did not identify a confirmed theft or breach.

The scanner also shifted infrastructure quickly, using more than 300 cloud addresses in one day and nearly 400 in one week. With addresses turning over at that rate, an IP blocklist alone would not address the pattern Cequence described.

What Cequence’s traffic figures measure

On one October day, the scanner reached 48% of the organizations protected by Cequence that had received any AI-bot traffic that day. The figure counts organizations in that specific group, not all organizations.

At several organizations, more than 80% of requests labeled as AI-bot traffic came from the impersonator. That percentage describes requests carrying an AI-bot label, not verified traffic from genuine agents. In a separate measure, customer-acting AI agents accounted for fewer than 1 in 4,000 requests in the financial-services organizations Cequence studied, even at the institution with the highest share. The two figures have different populations and denominators.

How organizations can check an AI bot’s identity

Cequence recommends verifying an agent through signed requests, published address ranges or a direct integration rather than trusting a name in a request. Teams can also compare a bot’s behavior with its stated role, keep secrets out of web-accessible files, and avoid exposing staging and administration sites publicly.