Jun 5, 2026·6 min read·61 visits
A dynamic sandbox bypass in Twig (CVE-2026-24425) allows remote code execution because runtime checks fail to propagate the active template's source context to filters like map or sort, allowing arbitrary PHP callables to run.
CVE-2026-24425 is a high-severity sandbox bypass vulnerability in the Twig template engine (affecting versions 2.16.x and 3.9.0 through 3.25.x). The flaw arises when dynamic sandboxing is enabled via a SourcePolicyInterface. During runtime checks of callback-accepting filters (such as map, sort, filter, and reduce), Twig queries the global sandbox status instead of the active template's context. This mismatch allows template authors to pass arbitrary PHP string callables to executive functions, resulting in unauthenticated or low-privilege Remote Code Execution (RCE) on the host.
The Twig template engine, widely used within the Symfony framework and modern PHP ecosystems, includes an optional sandbox mode designed to compile and execute untrusted templates securely. Many production deployments configure dynamic sandboxing using a SourcePolicyInterface. This interface evaluates whether a template must be sandboxed on a per-template basis, separating trusted system layouts from untrusted user-authored templates.
The attack surface is found in multi-tenant web applications, Content Management Systems (CMS), and software-as-a-service (SaaS) environments where untrusted users can submit custom template layouts. Because dynamic templates are marked sandboxed via the policy rather than a global switch, the execution engine must trace and apply isolation policies dynamically across all active contexts.
The vulnerability, identified as CVE-2026-24425, is a protection mechanism failure (CWE-693). When a sandbox is evaluated dynamically, the runtime validation engine fails to propagate the template's source context to specific array filter callback validations. This failure allows standard template authors to execute arbitrary PHP functions outside the sandbox environment.
To prevent remote code execution, Twig strictly limits the callback parameters allowed inside native template filters such as sort, filter, map, reduce, and find. Inside a sandbox environment, these filters must only accept PHP Closure objects (anonymous functions and short arrows) rather than arbitrary string callables. This prevention mechanism is governed by the checkArrow routine within the CoreExtension module.
public static function checkArrow(Environment $env, $arrow, $thing, $type)
{
if ($arrow instanceof \Closure) {
return;
}
if ($env->hasExtension(SandboxExtension::class) && $env->getExtension(SandboxExtension::class)->isSandboxed()) {
throw new RuntimeError(\sprintf('The callable passed to the "%s" %s must be a Closure in sandbox mode.', $thing, $type));
}
}The vulnerability exists because isSandboxed() is invoked without arguments. Calling isSandboxed() on the SandboxExtension without passing a specific template Source object queries the global sandbox state, which returns false under dynamic policy-based sandboxing.
Consequently, the validation check passes silently because the execution environment believes no sandbox is active. The engine then passes the unrestricted string callable directly to the underlying PHP runtime functions (such as uasort or array_map), leading to raw system execution.
The patch addresses the context propagation failure by modifying how the compiler and the extension filters access sandbox state. Filters now request the sandboxed state explicitly by utilizing the needs_is_sandboxed attribute.
In the patched version of src/Extension/CoreExtension.php, the core callback-accepting filters are updated to receive the template's active sandbox status as a boolean argument.
@@ -263,14 +263,14 @@ public function getFilters(): array
new TwigFilter('join', [self::class, 'join']),
new TwigFilter('split', [self::class, 'split'], ['needs_charset' => true]),
- new TwigFilter('sort', [self::class, 'sort'], ['needs_environment' => true]),
+ new TwigFilter('sort', [self::class, 'sort'], ['needs_environment' => true, 'needs_is_sandboxed' => true]),
new TwigFilter('merge', [self::class, 'merge']),
new TwigFilter('batch', [self::class, 'batch']),
new TwigFilter('column', [self::class, 'column']),
- new TwigFilter('filter', [self::class, 'filter'], ['needs_environment' => true]),
- new TwigFilter('map', [self::class, 'map'], ['needs_environment' => true]),
- new TwigFilter('reduce', [self::class, 'reduce'], ['needs_environment' => true]),
- new TwigFilter('find', [self::class, 'find'], ['needs_environment' => true]),
+ new TwigFilter('filter', [self::class, 'filter'], ['needs_environment' => true, 'needs_is_sandboxed' => true]),
+ new TwigFilter('map', [self::class, 'map'], ['needs_environment' => true, 'needs_is_sandboxed' => true]),
+ new TwigFilter('reduce', [self::class, 'reduce'], ['needs_environment' => true, 'needs_is_sandboxed' => true]),
+ new TwigFilter('find', [self::class, 'find'], ['needs_environment' => true, 'needs_is_sandboxed' => true]),The internal helper function signature drops its dependency on the Environment object, checking the newly supplied boolean directly.
@@ -2091,13 +2091,13 @@ public static function arrayEvery(Environment $env, $array, $arrow)
- public static function checkArrow(Environment $env, $arrow, $thing, $type)
+ public static function checkArrow(bool $isSandboxed, $arrow, $thing, $type)
{
if ($arrow instanceof \Closure) {
return;
}
- if ($env->hasExtension(SandboxExtension::class) && $env->getExtension(SandboxExtension::class)->isSandboxed()) {
+ if ($isSandboxed) {
throw new RuntimeError(\sprintf('The callable passed to the "%s" %s must be a Closure in sandbox mode.', $thing, $type));
}Additionally, legacy compiler pathways recover the calling template's context dynamically by walking the PHP stack trace. The twig_resolve_caller_source routine inspects the trace to ensure the sandbox configuration is robustly recovered and applied.
To exploit this vulnerability, an attacker must have privileges to author, upload, or modify a Twig template rendered under a dynamic SourcePolicyInterface sandbox configuration.
The attacker crafts a template payload containing a callback-accepting filter, such as map, sort, filter, or reduce. Instead of providing a secure anonymous function, the attacker passes a sensitive system callable string (e.g., 'system', 'passthru', 'shell_exec').
{# Attack vector using the map filter to invoke system execution #}
{{ ["id"]|map("system")|join }}During rendering, the compiler processes this block into PHP. Since the sandbox check is bypassed, the application maps the array containing the command argument ("id") directly to the PHP system() command. The host server executes the command and injects the output of the process into the rendered template context returned to the attacker.
The impact of CVE-2026-24425 is rated high (CVSS 8.8) because it leads directly to Remote Code Execution on the application server. Any host that accepts dynamically sandboxed templates from external users can be completely compromised by an attacker with standard template authoring permissions.
Successful execution grants shell access under the privileges of the web application user (e.g., www-data). Attackers can perform local file read operations, access database configuration strings, write web shells to public directories, or attempt local privilege escalation within containerized hosts.
Because the vulnerability bypasses Twig's sandboxing architecture, it undermines the security assumption of template isolation. Applications that rely on dynamic templates for multi-tenant customer dashboards or dynamic newsletter designs are highly susceptible to data leakage and host compromise.
The primary remediation strategy is upgrading the twig/twig dependency to a secure version. Organizations should audit their dependencies and apply the patch as soon as possible.
composer update twig/twigIf upgrading immediately is not feasible, developers should replace the dynamic SourcePolicyInterface dynamic sandbox architecture with a globally enabled sandbox. Global sandboxing is unaffected by this flaw because the global check evaluates to true, triggering the exception block on non-closure callbacks.
// Global configuration workaround
$policy = new \Twig\Sandbox\SecurityPolicy($allowedTags, $allowedFilters, $allowedMethods, $allowedProperties, $functions);
$sandbox = new \Twig\Extension\SandboxExtension($policy, true); // True enforces global isolation
$twig->addExtension($sandbox);Furthermore, web application firewalls (WAF) can be configured to filter template input blocks containing array filters combined with string constants. Implement standard security scanning to flag Twig templates containing patterns such as |map(, |sort(, or |filter( followed directly by a single or double-quoted string literal.
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H| Product | Affected Versions | Fixed Version |
|---|---|---|
Twig twigphp | >= 2.16.0, <= 2.16.1 | N/A |
Twig twigphp | >= 3.9.0, <= 3.25.1 | 3.26.0 |
| Attribute | Detail |
|---|---|
| CWE ID | CWE-693 |
| Attack Vector | Network |
| CVSS v3.1 | 8.8 |
| EPSS Score | 0.00114 |
| Exploit Status | Proof-of-Concept |
| CISA KEV Status | No |
The product does not use or incorrectly implements a protection mechanism, allowing attackers to bypass security restrictions.
A security vulnerability in @astrojs/netlify allows attackers to bypass remote image path restrictions by leveraging unescaped regular expression metacharacters. The integration adapter fails to sanitize developer-defined pathnames before interpolating them into a configuration JSON file consumed by Netlify's Edge Image CDN. This results in overly permissive matching behavior at the edge routing layer, enabling path-traversal and filter bypasses.
Prior to version 1.2.11, the LangChain LLM framework is affected by a Server-Side Request Forgery (SSRF) vulnerability inside its image token counting mechanism. Specifically, the ChatOpenAI.get_num_tokens_from_messages() method retrieves arbitrary image_url values from user prompts without validating the destination host or IP address. Attackers can exploit this issue to scan internal infrastructure, access local services, or harvest credentials from cloud metadata services.
A security vulnerability was identified in the Astro web framework's composable integration pipeline with Hono. Due to the structural coupling of CSRF origin validation exclusively within the `middleware()` primitive, applications assembling their routing pipeline manually could execute state-mutating actions before or entirely without origin validation. This flaw allows attackers to execute blind, write-only Cross-Site Request Forgery (CSRF) attacks against state-mutating Astro Actions or endpoints on behalf of authenticated users.
Multiple cross-site request forgery (CSRF) vulnerabilities in the HTTP Administration component in Cisco IOS 12.4 on the 871 Integrated Services Router allow remote attackers to execute arbitrary commands via crafted HTTP requests. This occurs because the web administrative server fails to validate request origins or use anti-CSRF tokens, allowing an attacker to abuse an active administrative session.
An unbounded resource allocation vulnerability exists in Guzzle's cookie parsing and storage engine. Prior to version 7.15.1, Guzzle did not restrict the number or size of cookies stored within the client-side CookieJar. This lack of boundary control allows a malicious server or an intermediary proxy to poison the cookie storage with oversized or excessive Set-Cookie headers. When the client subsequently targets sibling domains or legitimate services using the same cookie jar, it transmits exceptionally large Cookie headers, triggering upstream protocol violations and resulting in Denial of Service (HTTP 431 / HTTP 400 rejection).
An information disclosure vulnerability in guzzlehttp/guzzle allows host-only cookies to be incorrectly matched and sent to subdomains. Because Guzzle failed to track whether cookies were defined without a Domain attribute, it evaluated host-only cookies using standard domain-suffix rules, widening their scope and exposing sensitive session tokens to untrusted subdomains.