Someone is running mass vulnerability scans, spoofing AI bots like ClaudeBot
A new report unveils a surge in bot traffic, specifically malicious scans impersonating AI agents to target sensitive configuration files and credentials for AI coding tools. While some dismiss it as the 'water is wet' reality of the internet, the targeted nature and volume of these new AI-driven probes highlight an evolving threat landscape. Webmasters are now scrambling to defend against these sophisticated bots, which blend old scanning techniques with new AI-specific vulnerabilities.
The Lowdown
Known Agents has released its "Agentic Web Index," providing key metrics on over 5,000 websites to analyze bot traffic. The index details the rise of AI-related bots and flags a particular security concern: an active campaign of mass vulnerability scans spoofing AI bots like ClaudeBot.
- Overall bot traffic sees a slight decrease, but AI-related bot traffic, termed "Agentrification," has increased by 11%.
- The report categorizes AI agent activity into scraping (e.g., ClaudeBot for data collection), fetching (e.g., ChatGPT-User for RAG), search indexing (e.g., PetalBot for AI search results), and browsing (e.g., Google-Agent for autonomous tasks).
- Data indicates varying levels of
robots.txtcompliance, with some bots like Baiduspider and ShapBot being significant rule-breakers. - The primary alert is an "Active Threat: AI Bot Spoofing Campaign." Malicious actors are impersonating legitimate AI agents and performing widespread vulnerability scans.
- These scans specifically target credential and configuration paths for popular AI coding tools (e.g.,
.config/anthropic/credentials/default.json,.claude/settings.json,.envfiles). This suggests an attempt to exploit misconfigured or publicly exposed AI development environments.
The findings illustrate the growing complexity of web traffic, with an increasing proportion originating from diverse AI agents. More critically, they expose a new vector of cyber threat, where familiar scanning tactics are repurposed to exploit the burgeoning ecosystem of AI development tools.
The Gossip
Pervasive Probing: Nothing New Under the Sun
Many commenters expressed that mass vulnerability scanning is a constant, unavoidable reality of the internet, likening it to 'water is wet.' They argue that the only novelty is the specific user agents being spoofed, while the underlying behavior remains the same background noise that sysadmins have always dealt with. Others, including the author, pushed back, highlighting a statistically significant surge and the specific, new targets of these scans.
Defensive Measures & Mitigation Mastery
The discussion quickly shifted to practical strategies for detecting and blocking unwanted bot traffic. Suggestions included analyzing ASN ownership to block VPS providers, using firewall rules to block entire country IP ranges, and deploying specialized tools like Cloudflare Workers, Ubiquiti UniFi, or open-source honeypots for generating blocklists. There was also a debate on the cost-effectiveness of IP intelligence services.
Attribution & Authenticity Allegations
Commenters raised skepticism regarding the attribution of these scans to specific countries or even the claimed bot identities. They noted the ease of spoofing user agents and IP addresses, making it difficult to pinpoint the true origin or operator of malicious traffic. The author clarified that the report focuses on failed authentication attempts, indicating impersonation rather than direct attribution.
AI Tool Targets & Token Theft Theories
A key point of interest was the specific targeting of paths related to AI coding tools (e.g., config files, `.env` files, credentials). This suggests attackers are looking for exposed secrets to potentially gain access to AI agents or resources, with one theory being the theft of AI tokens to reduce operational costs. While some commenters saw this as part of a general vulnerability scan pattern, others emphasized the novel and concerning focus on the AI development ecosystem.