CVE-2026-56266

Multiple security vulnerabilities in the Crawl4AI Docker API server affecting endpoints for crawling, markdown/LLM extraction, screenshots, PDFs, webhooks, monitoring, JavaScript execution, and configuration.

Vulnerabilities

1. Arbitrary File Write via /screenshot and /pdf (CWE-22, CVSS 9.1)

The output_path parameter accepts arbitrary filesystem paths with no validation. An attacker can overwrite server files (DoS) or write to any appuser-writable location.

Fix: Added validate_output_path() restricting writes to CRAWL4AI_OUTPUT_DIR (/tmp/crawl4ai-outputs by default). Added Pydantic field_validator rejecting .. traversal sequences.

2. SSRF via Webhook URL (CWE-918, CVSS 8.6)

Webhook URLs in /crawl/job and /llm/job accept internal/private IPs with no validation, enabling Server-Side Request Forgery against cloud metadata endpoints (169.254.169.254), internal services, and Docker networks.

Fix: Added validate_webhook_url() with blocklist for RFC 1918, loopback, link-local, cloud metadata IPs and hostnames. Validation at both job submission and send time. Explicit follow_redirects=False.

3. Authentication Bypass on Monitor Endpoints (CWE-306, CVSS 6.5)

The monitor router was mounted without token_dep dependency, making all monitoring endpoints (including destructive ones like /monitor/actions/cleanup) accessible without authentication.

Fix: Added dependencies=[Depends(token_dep)] to monitor router. Added explicit token check on WebSocket /monitor/ws endpoint.

4. Stored XSS in Monitor Dashboard (CWE-79, CVSS 6.1)

URLs and error messages rendered in the monitor dashboard via innerHTML without escaping, enabling stored XSS via crafted crawl URLs.

Fix: Server-side html.escape() on URL and error storage. Client-side escapeHtml() wrapper on all innerHTML template injections.

5. Arbitrary JavaScript Execution via /execute_js (CWE-94, CVSS 8.1)

The /execute_js endpoint accepts and executes arbitrary JavaScript in the server's browser with --disable-web-security enabled, combining arbitrary JS execution with SSRF capability.

Fix: Disabled by default via CRAWL4AI_EXECUTE_JS_ENABLED env var. Added SSRF blocklist on destination URL. Removed --disable-web-security from default browser args.

6. Hardcoded JWT Secret Key (CWE-798, CVSS 9.8)

The JWT signing key defaults to "mysecret" in the public source code, allowing anyone to forge valid authentication tokens.

Fix: Removed default value. Added startup validation rejecting weak/short secrets. Auto-generates ephemeral key when JWT enabled but no key set.

7. SSRF via Direct Crawl Endpoints /crawl, /md, /llm (CWE-918, CVSS 8.6)

The primary crawl entry points (/crawl, /crawl/stream, /md, /llm) fetch arbitrary user-supplied URLs with no destination validation, enabling Server-Side Request Forgery against internal services, Docker networks, and cloud metadata endpoints (169.254.169.254). A blocklist that only inspects the literal hostname is additionally bypassable via IPv6-mapped IPv4 addresses (e.g. [::ffff:169.254.169.254], [::ffff:10.0.0.1]), which resolve to the blocked private/metadata ranges but evade a naive string check.

Fix: Added URL destination validation on all crawl/md/llm entry points, reusing the SSRF blocklist (RFC 1918, loopback, link-local, cloud-metadata IPs and hostnames). IPv6-mapped IPv4 addresses are normalized to their IPv4 form before the blocklist check, closing the mapping bypass. raw:// URLs are skipped. Validation applies at request entry, not only at fetch time.

Workarounds

1. Upgrade to the patched version (recommended)
2. Set CRAWL4AI_API_TOKEN to enable authentication
3. Set a strong SECRET_KEY (min 32 chars) if using JWT
4. Restrict network access to the Docker API

Credits

• Jeongbean Jeon - file write, SSRF, monitor auth bypass, stored XSS
• wulonchia - file write via output_path (independent report)
• by111 (August829) - hardcoded JWT, eval in /config/dump, /execute_js, hook sandbox escape
• secsys_codex - SSRF via /md, /crawl, /llm endpoints + IPv6-mapped IPv4 bypass (URL destination validation)
• Velayutham Selvaraj (LinkedIn) - SSRF via missing host validation in validate_url_scheme (independent report)
• IcySun & Yashon - SSRF, arbitrary file write, missing-auth-by-default, hook sandbox bypass via asyncio (independent report)

Input manipulates file paths to reach files outside the intended directory, such as configuration or credential files. Typical impact: unauthorized file read or write outside the intended directory.

CVE-2026-56266 has a CVSS score of 9.8 (Critical). The vector is network-reachable, no privileges required, and no user interaction. A CVSS score reflects the worst-case severity of the vulnerability, not your specific exposure. Whether this affects your application depends on whether the vulnerable code is present and reachable in your environment. A fixed version is available (0.8.7); upgrading removes the vulnerable code path.

Kodem intelligence

1. What is CVE-2026-56266? CVE-2026-56266 is a critical-severity path traversal vulnerability in crawl4ai (pip), affecting versions <= 0.8.6. It is fixed in 0.8.7. Input manipulates file paths to reach files outside the intended directory, such as configuration or credential files.
2. How severe is CVE-2026-56266? CVE-2026-56266 has a CVSS score of 9.8 (Critical). This score reflects the worst-case severity of the vulnerability, not your specific exposure. Whether it represents real risk in your environment depends on whether the vulnerable code is present and reachable.
3. Which versions of crawl4ai are affected by CVE-2026-56266? crawl4ai (pip) versions <= 0.8.6 is affected.
4. Is there a fix for CVE-2026-56266? Yes. CVE-2026-56266 is fixed in 0.8.7. Upgrade to this version or later.
5. Is CVE-2026-56266 exploitable, and should I be worried? Whether CVE-2026-56266 is exploitable in your environment depends on whether the vulnerable code is present and reachable. A CVSS score is a worst-case rating; it does not account for your specific deployment, configuration, or usage patterns. Kodem, an Intelligent Application Security platform, uses runtime intelligence to show which vulnerabilities actually execute in production, so you can focus on the ones that represent real risk. Get a demo
6. What actually determines whether CVE-2026-56266 is exploitable, and how bad it is? Exploitability and impact are not fixed properties of a CVE. They depend on runtime truth: whether the vulnerable code is present, reachable, and actually executes in your application. A high CVSS score on a dependency that never runs is not the same as real risk. Kodem, an Intelligent Application Security platform, uses runtime intelligence to reveal which vulnerabilities actually execute in production, so teams prioritize the ones that genuinely matter.
7. How do I fix CVE-2026-56266? Upgrade crawl4ai to 0.8.7 or later.