Description

Crawl4AI before 0.9.3 contains an uncontrolled resource consumption vulnerability in PDFContentScrapingStrategy that allows untrusted clients to cause denial of service. Attackers can select the PDF scraping strategy in POST requests to download large remote PDFs without size or page limits, exhausting disk, CPU, and bandwidth on shared workers.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "7.5",
  "pubDate": "2026-09-15T16:17:46.000Z",
  "pubdate": "2026-09-15T16:17:46.000Z",
  "executiveSummary": "Crawl4AI versions prior to 0.9.3 are susceptible to an uncontrolled resource consumption vulnerability, classified as a Denial of Service (DoS) flaw.\nThe vulnerability resides within the PDFContentScrapingStrategy component, which fails to enforce constraints on remote PDF retrieval processes.\nBy issuing crafted POST requests, unauthenticated remote attackers can force the application to download arbitrarily large PDF files.\nThis lack of validation leads to the exhaustion of critical system resources, including disk space, CPU cycles, and network bandwidth, specifically impacting shared workers.\nThe vulnerability carries significant risk for shared hosting environments or scaled microservices, as a single malicious actor can render the crawling service unavailable to legitimate users.\nExploitation requires no elevated privileges and relies entirely on the application's willingness to process attacker-supplied URLs without imposing size or pagination thresholds.",
  "technicalDetails": "The vulnerability is localized within the PDFContentScrapingStrategy class of the Crawl4AI framework, which is responsible for parsing and extracting content from remote PDF documents.\nThe root cause is a deficiency in input validation and resource management during the ingestion of external PDF resources. When the PDFContentScrapingStrategy is invoked via a POST request, the framework initiates an automated fetch operation for the provided remote document URL.\nCritically, the implementation lacks boundary enforcement, such as maximum file size limitations, document length limits (in pages), or timeout thresholds for the processing pipeline.\nAn attacker exploits this by providing a URL pointing to a malicious or exceptionally large PDF file. Upon receiving the request, the worker process proceeds to download the content to the local filesystem or memory, process the document, and render it for scraping purposes.\nBecause the system performs this synchronously or within shared worker threads without rate-limiting or resource quotas, multiple concurrent requests for large documents can quickly saturate the host system's hardware.\nThe attack flow follows a direct trajectory: 1) The attacker initiates an HTTP POST request targeting the Crawl4AI API endpoint; 2) The attacker specifies the PDFContentScrapingStrategy in the request payload; 3) The attacker provides a URL pointing to a high-volume payload (e.g., a multi-gigabyte PDF); 4) The worker process attempts to pull the resource, causing an immediate spike in network bandwidth utilization and disk I/O; 5) As the PDF is processed, CPU consumption increases significantly, and local storage may be exhausted if the application caches the raw file.\nThis behavior facilitates a persistent Denial of Service state, as the workers remain tied up processing the resource-heavy payload, effectively blocking legitimate crawl requests. Because the vulnerability exists in the ingestion layer, it is reachable by any user capable of interacting with the API, regardless of authentication state in non-secured deployments. The impact is primarily local resource starvation, which degrades the performance of the entire Crawl4AI infrastructure shared by the worker pool."
}