# CWE-95: Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection')

**Likelihood of exploit:** Medium · **KEV-linked**

> As of 2026-10-05, CWE-95 (Eval Injection) underlies 5 CVEs tracked by Threadlinqs, 2 of them in the CISA Known Exploited Vulnerabilities catalog, and is cited by 26 tracked threats. MITRE rates its likelihood of exploit as Medium.

**Last updated:** 2026-10-05

## What is CWE-95?

The product receives input from an upstream component, but it does not neutralize or incorrectly neutralizes code syntax before using the input in a dynamic evaluation call (e.g. "eval").

CWE-95 is a variant-level weakness in MITRE’s Common Weakness Enumeration, with a MITRE likelihood of exploit of Medium. Applicable platforms: Language: Java; Language: JavaScript; Language: Python; Language: Perl; Language: PHP; Language: Ruby.

_Source: [MITRE CWE](https://cwe.mitre.org/data/definitions/95.html) (CWE-95 definition, reproduced verbatim). Counts and linkage below are Threadlinqs data._

## Consequences

- **Confidentiality** — Read Files or Directories, Read Application Data. The injected code could access restricted data / files.
- **Access Control** — Bypass Protection Mechanism. In some cases, injectable code controls authentication; this may lead to a remote vulnerability.
- **Access Control** — Gain Privileges or Assume Identity. Injected code can access resources that the attacker is directly prevented from accessing.
- **Integrity, Confidentiality, Availability, Other** — Execute Unauthorized Code or Commands. Code injection attacks can lead to loss of data integrity in nearly all cases as the control-plane data injected is always incidental to data recall or writing. Additionally, code injection can often result in the execution of arbitrary code or at least modify what code can be executed.
- **Non-Repudiation** — Hide Activities. Often the actions performed by injected control code are unlogged.

_Source: MITRE CWE, common consequences._

## How CWE-95 is exploited in the wild

Threadlinqs maps 5 CVEs to CWE-95, published between 2024-07-01 and 2026-08-11. 2 are listed in CISA’s Known Exploited Vulnerabilities catalog, the authoritative record of exploitation in the wild. By CVSS v3 severity the set splits into 3 critical. The highest EPSS score in the set is 99.8% (CVE-2024-36401), the modelled probability of exploitation in the next 30 days. 26 tracked threats reference CWE-95 directly or through a CVE it covers; the most recent is “Chinese-Speaking 'Kapibala' Actor (Red Heron-Linked) Chains WordPress wp2shell, Zyxel GS1900, and Ubiquiti UniFi OS Flaws to Steal Government Data” (2026-09-22). Affected products concentrate in Adobe (1), Langflow (1), geoserver (1), among 5 vendors in total.

## Vulnerabilities (CVEs)

All 5 CVEs mapped to CWE-95, CISA KEV first, then by CVSS score.

- [CVE-2024-36401](https://intel.threadlinqs.com/cve/CVE-2024-36401) — CISA KEV · CVSS 9.8 critical · EPSS 99.8% · published 2024-07-01
- [CVE-2026-33017](https://intel.threadlinqs.com/cve/CVE-2026-33017) — CISA KEV · CVSS 9.8 critical · EPSS 5.6% · published 2026-03-20
- [CVE-2026-48317](https://intel.threadlinqs.com/cve/CVE-2026-48317) — CVSS 9.6 critical · published 2026-08-03
- [CVE-2026-53875](https://intel.threadlinqs.com/cve/CVE-2026-53875) — EPSS 0.4% · published 2026-06-17
- [CVE-2026-73248](https://intel.threadlinqs.com/cve/CVE-2026-73248) — EPSS 0.1% · published 2026-08-11

## Affected vendors

- [Adobe](https://intel.threadlinqs.com/vendors/adobe) — 1 CVE
- **Langflow** — 1 CVE
- **geoserver** — 1 CVE
- **kovidgoyal** — 1 CVE
- [picklescan](https://intel.threadlinqs.com/vendors/picklescan) — 1 CVE

## Threat activity

26 tracked threats cite CWE-95; the 25 most recent are listed.

- [Chinese-Speaking 'Kapibala' Actor (Red Heron-Linked) Chains WordPress wp2shell, Zyxel GS1900, and Ubiquiti UniFi OS Flaws to Steal Government Data](https://intel.threadlinqs.com/threat/TL-2026-2619) — CRITICAL · 2026-09-22
- [Unauthenticated AWS API Gateway + Over-Permissioned Lambda: Credential Extraction Attack Chain](https://intel.threadlinqs.com/threat/TL-2026-2601) — HIGH · 2026-09-21
- [Endor Labs Discloses 14 Critical/High Vulnerabilities Across Seven AI Orchestration Platforms (NocoBase, Flowise, Langflow, Dify, Activepieces, Kestra, Apache Airflow)](https://intel.threadlinqs.com/threat/TL-2026-2438) — CRITICAL · 2026-09-10
- [Adobe and Nvidia Patch Dozens of Vulnerabilities Across Multiple Products, Including Two Critical Flaws in Nvidia's NemoClaw AI Agent Stack and a CVSS 10.0 Adobe Campaign Classic Chain](https://intel.threadlinqs.com/threat/TL-2026-2152) — CRITICAL · 2026-08-26
- [AI-Powered Attacks Targeting Siemens S7 Series PLCs in U.S. Critical Infrastructure](https://intel.threadlinqs.com/threat/TL-2026-2076) — CRITICAL · 2026-08-19
- [Adobe Patches Critical RCE Flaws in ColdFusion, Campaign Classic, and Commerce (CVE-2026-48362, CVSS 10.0)](https://intel.threadlinqs.com/threat/TL-2026-1985) — CRITICAL · 2026-08-11
- [Adobe Campaign Classic Critical Incorrect Authorization Flaw Enables Unauthenticated Remote Code Execution (CVE-2026-48449) Paired With SQL Injection Memory/File Disclosure (CVE-2026-48448)](https://intel.threadlinqs.com/threat/TL-2026-1790) — CRITICAL · 2026-07-31
- [CISA Orders Federal Agencies to Patch Actively Exploited Langflow RCE Flaw (CVE-2026-0770)](https://intel.threadlinqs.com/threat/TL-2026-1618) — CRITICAL · 2026-07-22
- [StrikeShark Campaign: SharkLoader Dropper Targets Governments and Software Developers via N-Day Exploits and Trojanized Installers to Deploy Cobalt Strike](https://intel.threadlinqs.com/threat/TL-2026-1237) — HIGH · 2026-07-11
- [JADEPUFFER: AI Agent Exploits Langflow RCE (CVE-2025-3248) to Automate Database Ransomware/Extortion Attack](https://intel.threadlinqs.com/threat/TL-2026-1117) — CRITICAL · 2026-07-05
- [SharkLoader Malware Campaign Uses Fake Cisco AnyConnect and Google Update Installers to Deploy Cobalt Strike](https://intel.threadlinqs.com/threat/TL-2026-1101) — HIGH · 2026-07-03
- [JADEPUFFER: First End-to-End Agentic Ransomware Attack Exploiting Langflow (CVE-2025-3248) and Nacos (CVE-2021-29441)](https://intel.threadlinqs.com/threat/TL-2026-1083) — CRITICAL · 2026-07-02
- [Langflow CVE-2026-33017 Unauthenticated RCE Actively Exploited for Monero Mining (lambsys)](https://intel.threadlinqs.com/threat/TL-2026-1002) — CRITICAL · 2026-06-30
- [SEO Poisoning Supply Chain Campaign Distributing Akira Ransomware via Trojanized Enterprise Software](https://intel.threadlinqs.com/threat/TL-2026-1004) — CRITICAL · 2026-06-30
- [Mistic Windows Backdoor - In-Memory Code Execution via DLL Sideloading](https://intel.threadlinqs.com/threat/TL-2026-1017) — CRITICAL · 2026-06-30
- [JadeSnow: Hijacked npm/Go Supply Chain Attack with VSCode Exploitation and Blockchain Dead Drops](https://intel.threadlinqs.com/threat/TL-2026-0987) — CRITICAL · 2026-06-28
- [Dropping Elephant Malware Campaign - China-Themed Loader Chain for Initial Access and Payload Delivery](https://intel.threadlinqs.com/threat/TL-2026-0989) — CRITICAL · 2026-06-28
- [SharkLoader Malware Deploys Cobalt Strike Beacon via DLL Side-Loading in StrikeShark Campaign](https://intel.threadlinqs.com/threat/TL-2026-0961) — HIGH · 2026-06-27
- [Q1 2026 Ransomware Landscape: Qilin Dominance, LockBit 5.0 Comeback, and FortiGate (CVE-2024-55591) / Oracle EBS (CVE-2025-61882) Mass Exploitation](https://intel.threadlinqs.com/threat/TL-2026-0758) — CRITICAL · 2026-06-10
- [Lazarus Group npm Brandjacking Campaign — buffer-utilities Multi-Stage Staging Framework (sonatype-2026-003558)](https://intel.threadlinqs.com/threat/TL-2026-0726) — HIGH · 2026-06-09
- [Everest Forms Pro WordPress Plugin CVE-2026-3300 — Unauthenticated RCE via Calculation Addon process_filter() eval() Injection (CVSS 9.8, Active Exploitation)](https://intel.threadlinqs.com/threat/TL-2026-0689) — CRITICAL · 2026-06-06
- [NATS-as-C2: KeyHunter Distributed Worker Botnet Harvests Cloud Credentials and AI API Keys via Langflow RCE (CVE-2026-33017)](https://intel.threadlinqs.com/threat/TL-2026-0514) — HIGH · 2026-05-14
- [Microsoft Semantic Kernel Prompt-Injection-to-RCE & Arbitrary File Write — CVE-2026-26030 (InMemoryVectorStore eval) + CVE-2026-25592 (SessionsPythonPlugin)](https://intel.threadlinqs.com/threat/TL-2026-0489) — CRITICAL · 2026-05-09
- [CVE-2026-33017: Langflow Unauthenticated RCE via Public Flow Build Endpoint — Active Exploitation Within 20 Hours](https://intel.threadlinqs.com/threat/TL-2026-0250) — CRITICAL · 2026-03-19
- [Ivanti EPMM Dual-CVE Unauthenticated RCE Chain (CVE-2026-1281 + CVE-2026-1340) — CVSS 9.8, CISA KEV, Dutch Government Breached, Bulletproof Hosting IAB, Sleeper Webshells, 28K+ Attacking IPs](https://intel.threadlinqs.com/threat/TL-2026-0121) — CRITICAL · 2026-02-16

## Mitigations

- **Architecture and Design, Implementation**: If possible, refactor your code so that it does not need to use eval() at all.
- **Implementation / Input Validation**: Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected…
- **Implementation**: Inputs should be decoded and canonicalized to the application's current internal representation before being validated (CWE-180, CWE-181). Make sure that your application does not inadvertently decode the same input twice (CWE-174). Such errors could be used to bypass allowlist schemes by introducing dangerous inputs after they have been checked. Use libraries such as the OWASP ESAPI Canonicalization control. Consider performing repeated canonicalization until your input does not change any more. This will avoid double-decoding and similar scenarios, but it might inadvertently modify inputs that are allowed to contain properly-encoded dangerous content.
- **Implementation**: For Python programs, it is frequently encouraged to use the ast.literal_eval() function instead of eval, since it is intentionally designed to avoid executing code. However, an adversary could still cause excessive memory or stack consumption via deeply nested structures [REF-1372], so the python documentation discourages use of ast.literal_eval() on untrusted data [REF-1373].

_Source: MITRE CWE, potential mitigations._

## Detection methods (MITRE CWE)

- **Automated Static Analysis** (effectiveness: High): Automated static analysis, commonly referred to as Static Application Security Testing (SAST), can find some instances of this weakness by analyzing source code (or binary/compiled code) without having to execute it. Typically, this is done by building a model of data flow and control flow, then searching for potentially-vulnerable patterns that connect "sources" (origins of input) with "sinks" (destinations where the data interacts with external components, a lower layer such as the OS, etc.)

_Source: MITRE CWE, detection methods. Threadlinqs detection rules for the threats above are Blue tier and higher._

## Related weaknesses

- [CWE-94 Code Injection](https://intel.threadlinqs.com/cwe/CWE-94)

Canonical: https://intel.threadlinqs.com/cwe/CWE-95
Source definition: https://cwe.mitre.org/data/definitions/95.html
Detection rules and IOCs for threats exploiting CWE-95 via the Threadlinqs MCP server (Purple tier): https://intel.threadlinqs.com/mcp