Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-112363 Overview
CVE-2026-112363 is a Server-Side Request Forgery (SSRF) vulnerability [CWE-918] affecting Crawl4AI versions 0.8.7 through 0.9.5. The flaw resides in the Docker API server, where destination validation can be bypassed through differences in how URL userinfo components are parsed. Unauthenticated attackers can submit crafted URLs with the fetch_ssl_certificate=true parameter, causing SSLCertificate.from_url to establish TLS connections to internal hosts on port 443. This behavior enables probing of internal network reachability and leaks TLS certificate details from systems that should not be exposed.
Critical Impact
Unauthenticated remote attackers can bypass SSRF protections to probe internal infrastructure and exfiltrate internal TLS certificate metadata through the Crawl4AI Docker API.
Affected Products
- Crawl4AI 0.8.7 through 0.9.5
- Crawl4AI Docker API server deployments
- Applications integrating SSLCertificate.from_url from the affected versions
Discovery Timeline
- 2026-10-19 - CVE-2026-112363 published to NVD
- 2026-10-19 - Last updated in NVD database
Technical Details for CVE-2026-112363
Vulnerability Analysis
The vulnerability stems from inconsistent URL parsing between the Crawl4AI Docker API's destination allowlist logic and the downstream SSL fetch routine. When the API validates a URL, the host component is interpreted one way; when SSLCertificate.from_url resolves and connects to that same URL, it interprets the userinfo portion differently and extracts a different host. Attackers exploit this discrepancy by embedding an allowed host in the userinfo field while placing an internal target in the actual authority component.
When a client sends a request with fetch_ssl_certificate=true, the server opens a TLS session to the attacker-chosen internal host on port 443. The handshake returns certificate metadata to the attacker, revealing whether the host is reachable and disclosing certificate subject, issuer, serial, and SAN entries. This provides reconnaissance of internal services, cloud metadata hosts, and administrative endpoints without authentication.
Root Cause
The root cause is a parser differential between the URL validation layer and SSLCertificate.from_url in crawl4ai/ssl_certificate.py. The Docker utility functions in deploy/docker/utils.py do not normalize the URL before enforcement, allowing userinfo tokens such as allowed-host@internal-host to be interpreted as different authorities by each component.
Attack Vector
Attack delivery occurs over the network through an unauthenticated POST request to the Crawl4AI Docker API endpoint. The attacker supplies a URL containing userinfo that satisfies destination checks while the underlying TLS client connects to an internal address. No user interaction or credentials are required. The returned certificate payload confirms whether the target host responded on port 443, allowing systematic enumeration of internal services. See the VulnCheck SSRF Bypass Advisory and the GitHub Pull Request #2304 for technical details and the proposed fix.
Detection Methods for CVE-2026-112363
Indicators of Compromise
- Inbound requests to the Crawl4AI Docker API containing fetch_ssl_certificate=true paired with URLs that include @ in the userinfo portion.
- Outbound TLS connections from Crawl4AI containers to RFC1918 ranges, link-local addresses such as 169.254.169.254, or internal service hostnames on port 443.
- Repeated short-lived TLS handshakes from Crawl4AI workers targeting sequential internal IP addresses.
Detection Strategies
- Inspect API access logs for URL parameters where the host before @ and the host after @ differ.
- Correlate Crawl4AI process network telemetry with destination IPs that fall outside expected crawl targets.
- Alert on TLS SNI values resolving to internal zones originating from the Crawl4AI service account.
Monitoring Recommendations
- Enable egress logging on containers running Crawl4AI and forward DNS and connection events to a central analytics pipeline.
- Baseline normal destination domains for the Docker API and flag deviations toward internal CIDR blocks.
- Monitor process-level network activity from crawl4ai Python workers for connections to cloud metadata endpoints.
How to Mitigate CVE-2026-112363
Immediate Actions Required
- Restrict network access to the Crawl4AI Docker API so only trusted clients can reach it, and require authentication at the gateway.
- Block egress from Crawl4AI containers to RFC1918 ranges, loopback, and cloud metadata endpoints such as 169.254.169.254.
- Disable the fetch_ssl_certificate feature in production configurations where it is not required.
Patch Information
A fix is tracked in GitHub Pull Request #2304, which normalizes URL parsing before destination enforcement so userinfo cannot be used to bypass host validation. Operators running versions 0.8.7 through 0.9.5 should upgrade to a release containing this patch. Review the GitHub SSL Certificate Module and GitHub Async Crawler Strategy for the affected code paths.
Workarounds
- Deploy the Crawl4AI Docker API behind a reverse proxy that strips URLs containing userinfo components before forwarding requests.
- Enforce an outbound allowlist at the host firewall limiting destinations to public crawl targets only.
- Reject API requests where the submitted URL fails strict host == netloc validation after canonicalization.
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

