CVE-2026-91941 in Crawl4AI
Summary
by MITRE • 09/15/2026
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.
Statistical analysis made it clear that VulDB provides the best quality for vulnerability data.
Analysis
by VulDB Data Team • 09/16/2026
The identified vulnerability represents a critical flaw within Crawl4AI prior to version 0.9.3, specifically residing in the implementation of the PDFContentScrapingStrategy component. This security issue is classified as an uncontrolled resource consumption vulnerability, which fundamentally stems from insufficient validation and constraint enforcement when processing external resources. The core technical failure lies in the application's inability to impose strict limits on the size or page count of Portable Document Format files retrieved during scraping operations. When a user initiates a POST request that explicitly selects this specific PDF scraping strategy, the system proceeds to download the target document without performing any preliminary checks regarding file dimensions or structural complexity. This lack of boundary control allows an untrusted client to manipulate the input parameters to trigger the retrieval of excessively large files from remote servers.
From an operational perspective, this flaw enables a straightforward denial-of-service attack vector against shared worker environments. Because Crawl4AI is often deployed in multi-tenant or high-concurrency architectures where resources are pooled among multiple users or processes, the consumption of excessive disk space, central processing unit cycles, and network bandwidth by a single malicious request can have cascading effects on system stability. An attacker can exploit this weakness by submitting requests for massive PDF files that may contain numerous pages, embedded high-resolution images, or complex vector graphics designed to maximize parsing overhead. The resulting resource exhaustion leads to the degradation of service availability for legitimate users and potentially causes complete system unresponsiveness if the shared workers are overwhelmed beyond their capacity thresholds.
This vulnerability aligns with Common Weakness Enumeration identifier CWE-400, which describes situations where a software application does not properly control the allocation and maintenance of a limited resource, thereby allowing an attacker to consume excessive amounts of that resource. Furthermore, in the context of the MITRE ATT&CK framework, this behavior corresponds to techniques associated with Resource Hijacking or Denial of Service, specifically under tactics aimed at impacting availability rather than confidentiality or integrity. The attack requires no authentication if the endpoint is publicly accessible and relies on the social engineering aspect of convincing a user or automated system to submit a crafted POST request targeting the vulnerable scraping strategy.
To mitigate this risk, immediate remediation involves upgrading Crawl4AI to version 0.9.3 or later, where these constraints have been addressed by developers through the implementation of strict size limits and page count caps for PDF processing operations. In environments where an upgrade is not immediately feasible, defensive measures should include configuring reverse proxies or web application firewalls to inspect POST payloads and block requests that attempt to access large remote files without proper authorization headers. Additionally, implementing rate limiting on endpoints exposed to untrusted clients can help mitigate the impact of such attacks by throttling the frequency of resource-intensive operations. It is also advisable to isolate scraping workers in sandboxed environments with strict memory and disk quotas to prevent a single malicious request from compromising the entire host infrastructure.