Skip to content
Submit incident
AII-1001-11621
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.

OpenClaw ACP missing prompt-size checks in stdio bridge

OpenClaw's ACP bridge lacked prompt-size validation in its local stdio interface, allowing very large inputs to create oversized payloads that could reduce responsiveness and increase model costs. The vulnerability was patched in version 2026.2.18 with a 2 MiB prompt limit and improved size checks.

20 February 2026 Β· OpenClaw Β· Security and cyberattacks
20 February 2026
AII-1001-11621View source β†—
Quick facts
Organization
OpenClaw (subject)
Category
Security and cyberattacks
Source
github.com
Date
20 February 2026
Drafted by
Raidu Incident Detection

What happened

The ACP bridge in OpenClaw versions 2026.2.17 and earlier lacked prompt-size checks in the local stdio bridge, allowing very large prompt text blocks to be accepted and assembled into oversized payloads before forwarding to chat.send.

The vulnerability affected local ACP clients such as IDE integrations that send unusually large inputs through the stdio bridge.

Reported impact

Local ACP sessions could become less responsive when very large prompts were submitted.

Oversized text forwarded to the model could result in larger-than-expected model usage and costs.

Index analysis

The vulnerability involved missing input validation (CWE-20) and uncontrolled resource consumption (CWE-400) in the ACP event-mapper and translator components.

The local stdio transport meant no privilege escalation or direct remote attack path existed in the default ACP model.

Governance controls

OpenClaw patched the vulnerability in version 2026.2.18 by enforcing a 2 MiB prompt-text limit before concatenation, accounting for inter-block newline separator bytes during pre-concatenation size checks, and maintaining final outbound message-size validation before chat.send.

The fix also avoided stale active-run session state when oversized prompts were rejected and added regression tests for oversize rejection and active-run cleanup.

Classification

Organization
OpenClaw (subject)
Category
Security and cyberattacks
Harm status
Potential harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Economic or property
Harmed
Consumers
AI systems
OpenClaw

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

github.com
https://github.com/openclaw/openclaw/security/advisories/GHSA-cxpw-2g23-2vgw
Aug 2026Claude AI agent discovers gym booking system vulnerability, cancels another user's reservationWell documentedSecurity and cyberattacksThe IndependentDiaspora Digital Media (DDM News) - Nigeria Breaking News, Africa and World News and Updates -The Indian Express+3Jun 2026Five malicious skills removed from OpenClaw ClawHub marketplaceConfirmedSecurity and cyberattacksUnit42Dark ReadingJun 2026OpenClaw AI agent vulnerabilities enable malicious skill packages to steal user dataUnder investigationSecurity and cyberattacksenglish.news.cnChina NewsApr 2026AI Agent Exploited Gym Booking API Authorization Flaw and Removed Member From WaitlistReportedSecurity and cyberattacksabc.net.auaffinda.comApr 2026Open-source AI agents expose security vulnerabilities in ChinaReportedSecurity and cyberattacksδΈœζ–Ήθ΄’ε―Œη½‘China News