GreyNoise Finds AI Crawler Impersonation Used to Scan for Exposed Credentials

Summary

GreyNoise identified 824 IPs spoofing AI-crawler identities to scan for exposed credentials and configuration files. None matched published vendor ranges; it could not confirm that any files were accessed or secrets stolen.

Key points

• Between July 28 and August 23, 2026, traffic using six AI crawler names shared a single HTTP client fingerprint and came from the same 824 IP addresses.
• The addresses did not match published IP ranges for Anthropic, OpenAI, Google, Perplexity, or Amazon; the traffic also did not request /robots.txt.
• The scanners requested exposed credential and configuration paths, including /.env, /.env.production, /.env.bak, and /.aws/credentials.
• The shared fingerprint had carried more than 1,500 user-agent strings over the prior 90 days, and the 824 addresses spanned 795 separate /24 networks.
• GreyNoise also found forged Amazon crawler names using user-agent strings Amazon does not document.
• GreyNoise could not confirm whether requests returned files or whether any organization was affected, and did not identify who was behind the activity.
• GreyNoise published the IP addresses, targeted paths, and vendor address lists so site owners can check their logs.

MITRE ATT&CK