### Summary 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](https://github.com/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](https://www.linkedin.com/in/velayuthamselvaraj)) - 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)
### Summary 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](https://github.com/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](https://www.linkedin.com/in/velayuthamselvaraj)) - 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)
Update crawl4ai to 0.8.7 if you use the affected versions. Test the change in a non-production environment first.
Local check
hol-guard supply-chain scanCrawl4AI: Multiple Docker API Vulnerabilities - File Write, SSRF, Auth Bypass, XSS, JS Execution affects crawl4ai (pip). Severity is critical. ### Summary 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](https://github.com/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](https://www.linkedin.com/in/velayuthamselvaraj)) - 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)
AI coding agents often install or upgrade packages automatically in pip. A critical vulnerability in a dependency can be pulled into a project through a normal install or update without a human reviewing the change, expanding the blast radius from a single package to every agent workspace that depends on it.
| Package | Affected range | Fixed version |
|---|---|---|
| crawl4aipip | <=0.8.6 | 0.8.7 |
Fixed versions are reported by the source feed; confirm compatibility before updating.
Reported by GitHub Security Advisories (ghsa).
HOL Guard can help your team review package activity against supported protection paths.
Explore HOL GuardUpdate crawl4ai to 0.8.7 if you use the affected versions. Test the change in a non-production environment first.
Local check
hol-guard supply-chain scanCrawl4AI: Multiple Docker API Vulnerabilities - File Write, SSRF, Auth Bypass, XSS, JS Execution affects crawl4ai (pip). Severity is critical. ### Summary 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](https://github.com/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](https://www.linkedin.com/in/velayuthamselvaraj)) - 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)
AI coding agents often install or upgrade packages automatically in pip. A critical vulnerability in a dependency can be pulled into a project through a normal install or update without a human reviewing the change, expanding the blast radius from a single package to every agent workspace that depends on it.
| Package | Affected range | Fixed version |
|---|---|---|
| crawl4aipip | <=0.8.6 | 0.8.7 |
Fixed versions are reported by the source feed; confirm compatibility before updating.
Reported by GitHub Security Advisories (ghsa).
HOL Guard can help your team review package activity against supported protection paths.
Explore HOL Guard