Threat · curated 17 Jul 2026

NVD - CVE-2026-30741

Dossier

Coverage timeline

17 Jul 2026nist.gov

Single-source advisory — first reported, latest, and curated coincide.

Why it matters

CVE-2026-30741 shows prompt injection escalating directly to arbitrary code execution in an AI agent platform, meaning a maliciously crafted prompt can fully compromise the host running the agent.

CVE-2026-30741 is a remote code execution vulnerability in OpenClaw Agent Platform v2026.2.6 that allows attackers to execute arbitrary code via a request-side prompt injection attack. CISA-ADP rated it CVSS 3.1 base 9.8 (Critical), classified as CWE-94 code injection, with public exploit references including a GitHub PoC repository.

vuln-research

Summary

CVE-2026-30741 is a critical remote code execution vulnerability in the OpenClaw Agent Platform (v2026.2.6 and earlier) that allows an attacker to execute arbitrary code through a Request-Side prompt injection attack. CISA-ADP scored it CVSS 3.1 9.8 (Critical) and classified the root cause as CWE-94 code injection.[1]

The disclosure is supported by publicly available proof-of-concept material—a GitHub exploit repository referenced by MITRE and a demonstration video—rather than reports of coordinated in-the-wild exploitation; CISA's SSVC assessment records exploitation as 'none' while noting the attack is automatable with total technical impact.[1][2]

The accompanying demonstration illustrates that abusing third-party API keys and injecting prompts can coerce the agent into leaking sensitive data such as wallet mnemonic seed phrases, underscoring the practical impact of prompt-injection flaws in agent platforms.[2]

Attack chain

  1. Injection: An attacker submits a crafted Request-Side prompt injection payload to the OpenClaw Agent Platform.[1]
  2. Code execution: The injected content is improperly treated as code (CWE-94), resulting in arbitrary/remote code execution on the platform.[1]
  3. Impact / data theft: Demonstration material shows the technique combined with third-party API key hijacking to inject prompts and extract sensitive secrets such as wallet mnemonic seed phrases.[2]

Disclosure timeline

DateEvent
2026-02-05Demonstration video published on Bilibili showing third-party API key hijacking and prompt injection to obtain a mnemonic seed phrase.[2]
2026-03-11CVE-2026-30741 received from MITRE and published on NVD with description and references.[1]
2026-03-12CISA-ADP added CVSS 3.1 vector (9.8 Critical) and CWE-94 classification; SSVC timestamp recorded.[1]
2026-03-17NIST performed initial analysis, adding CPE configuration and reference types.[1]
2026-06-17CVE last modified by CISA-ADP and MITRE (SSVC and affected metadata updates).[1]

How it works

The vulnerability is a Request-Side prompt injection flaw in the OpenClaw Agent Platform. Malicious input carried in a request is improperly incorporated into generated code (CWE-94, Improper Control of Generation of Code), permitting an unauthenticated remote attacker to execute arbitrary code. The CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) indicates network attack vector, low complexity, no privileges or user interaction required, with high confidentiality, integrity, and availability impact.[1]

Public demonstration material shows the injection path combined with hijacking of third-party API keys, enabling the injected prompt to extract sensitive secrets such as cryptocurrency wallet mnemonic seed phrases.[2]

Affected versions and patch status

ProductAffectedPatch status
OpenClaw Agent Platform (openclaw:openclaw, Node.js)All versions up to and including 2026.2.6No fixed version or patch is identified in the available evidence.[1]

Key takeaways

  • Prompt injection against AI agent platforms can escalate to full remote code execution (CWE-94), as evidenced by the CVSS 9.8 rating for CVE-2026-30741.[1]
  • Although exploitation is currently assessed as 'none' in the wild, working proof-of-concept code and a public demonstration exist, and CISA notes the attack is automatable—raising the risk of near-term weaponization.[1][2]
  • Handing agents third-party API keys and sensitive secrets is dangerous, since injected prompts have been demonstrated to leak wallet mnemonic seed phrases.[2]

Defensive actions

  • Restrict and validate the use of third-party API keys with OpenClaw agents and avoid supplying secrets the agent could be induced to disclose.: Demonstration material shows third-party API key hijacking combined with prompt injection to exfiltrate sensitive data such as wallet mnemonic seed phrases.[2]
  • Treat all request-side input to the OpenClaw Agent Platform as untrusted and monitor for exploitation attempts, prioritizing an upgrade beyond 2026.2.6 once available.: The flaw allows unauthenticated remote code execution via prompt injection (CVSS 9.8) and CISA notes the attack is automatable, so exploitation attempts are low-effort and high-impact.[1]