OpenClaw
Named by 61 kept items in this workspace.
32 items name it under affected products, 14 under classification.technologies, 10 under tools, 3 under classification.key entities and 2 under threat actors.
The items that name it 61
personal AI agent harness
AI coding agent among the 18 covered by Dev Machine Guard.
subject of 27 advisory submissions
Agent platform/gateway that the compromised npm plugin integrates with for memory recall and storage
Used in distillation campaigns to harvest chain-of-thought, per Anthropic
Telegram/URL previews
agent harness built as gateway with registered nodes exposing system commands, camera, contacts
Open-source AI agent framework; subject of the AI SAST analysis, found to have 7 exploitable vulnerabilities spanning HTTP route handlers, business logic, tool dispatch layers, and lower-level implementation functions
versions prior to 2026.2.14; non-sandboxed / non-containerized deployments without sandboxRoot configured, running via 'openclaw gateway run' natively outside Docker
Tool Troy Hunt used to analyze the Carhartt breach data.
AI agent; before_tool_call callback monitored via numbat's generated plugin, produced network.indicator event for cloud-metadata browser request
Target of 'Salami attack' stealthy collusive memory poisoning, achieving 81.3% fragments saved and 75.0% attack success across 48 scenarios and two backbone models
agent Pinchy leaked mock AWS keys and CRM data in phishing tests
Open-source AI agent harness used to interact with web systems on behalf of a user; described as a 'notoriously loose harness when it comes to security'
Agent platform that adopted the SKILL.md skill format; target platform for SecureClaw tool
agent tool configured to use the model via a local llama.cpp server
Autonomous AI agent (using Anthropic's Claude) used to help a user manage a gym class waitlist; discovered and exploited an API authorization gap
Long-running Node.js service connecting an LLM to a Mac Mini and messaging apps; supplies gateway, channels, context assembly, persistent memory, skills, and the agentic loop
150 OpenClaw task directories in SeClaw-Bench dataset used as evaluation targets for agent security tasks
AI agent platform covered by agent-audit, AI-Infra-Guard, trentclaw, clawsec, SlowMist Agent Security
LLM agent runtime/environment
Open source personal AI assistant that lives on a machine or in the cloud, talks over WhatsApp or Telegram, clears inbox, manages calendar, browses the web, runs shell commands, and writes its own plugins; affected by CVE-2026-25253 (ClawBleed)
AI coding agent framework; SuperClaw attacks/audits OpenClaw instances (e.g. via ws://127.0.0.1:18789), which often run with broad tool access and can ingest untrusted, adversarial content from agent networks like Moltbook.
Open-source agentic framework launched January 2026; accumulated over 336,000 GitHub stars and spawned more than 2,100 agents within 48 hours of release; security audit identified 512 vulnerabilities including CVE-2026-25253; over 1,800 exposed instances leaking API keys and credentials within the f
Malware referenced in public victim write-ups sharing same wallet addresses, Aptos fallback identifiers, XOR keys, and config-file injection pattern
Supported AI coding tool target; framework emits AGENTS.md + .agents/skills/ + openclaw.json
Platform associated with a dedicated skill registry (ClawHub) that has been a growing target for threat actors creating malicious skills to exfiltrate credentials, execute arbitrary code, and more.
agentic AI runtime governed by DefenseClaw
Agentic AI codebase analyzed across MAESTRO's 7 layers (Foundation Models, Data Operations, Agent Frameworks, Deployment & Infrastructure, Evaluation & Observability, Security & Compliance, Agent Ecosystem)
Open-source AI assistant project impersonated by fake Windows and macOS installers; original release November 2025 as Clawdbot, rebranded Moltbot late January 2026, then OpenClaw three days later
indirect prompt injection lab target
Gateway accepts WebSocket connections from localhost, allowing any website to take control of AI agent
SKILL.md YAML skill format; patched versions 2026.2.13 (log poisoning) and 2026.2.25 (ClawJacked).
AI assistant framework/agent platform used to build Fiu
Open-source community whose vulnerabilities were fixed with help from Tencent Zhuque Lab; also integration target via aig-scanner skill and ClawHub
AI agent framework at the center of the ClawHub supply chain incident and a cluster of March 2026 security papers
In every workspace you can read 92
AI
Security
All technologies in this workspace.