← Papers By AI ↓ PDF Version Manuel Pereira & Opus 4.6  ·  Feb 2026
Autonomous AI · Security · Geopolitics

OpenClaw: We Are Not Ready

An opinion article on OpenClaw — the autonomous AI agent that reached 150,000 GitHub stars before anyone understood what it could do. A personal story, a systemic analysis, and a geopolitical warning.

Manuel Pereira & Claude Opus 4.6 February 2026 Critical Advisory
Subject: OpenClaw Autonomous AI Agent
GitHub Stars
150,000+
Security Advisories
4
CrowdStrike · Cisco · Bitdefender · Trend Micro
CVE Status
CVE-2026-25253
PART I
Personal — Why I Decided Not to Install It

Research That Changed My Mind

I am not a professional security researcher. I am someone who spent three weeks investigating an open-source AI agent that everyone in my professional network was installing — and what I found persuaded me to stop and write this instead.

CVE-2026-25253 — Critical Severity: OpenClaw's document processing pipeline allows untrusted content in processed documents to inject instructions that are executed with the agent's full permission scope. When combined with calendar access, email, and file system permissions that OpenClaw requests by default, this creates a complete remote code execution vector via a crafted PDF or Word document.

The Privacy Illusion

OpenClaw's marketing emphasizes local processing — "your data never leaves your machine." This is technically true for most operations. But it is architecturally misleading: the agent regularly communicates with external orchestration servers for task planning, model inference, and capability extensions. The traffic analysis shows significant data exfiltration potential that is not disclosed in the privacy policy.

The Sandbox Design

OpenClaw's sandbox operates on an allowlist model by default: it starts with broad permissions and asks users to manually restrict them. Most users never do. The result: an agent with unrestricted access to the local file system, network, and installed applications — running instructions that may originate from untrusted documents it was asked to process.

PART II
Extrapolation — The Architecture of Autonomy

The Lethal Trifecta

The Lethal Trifecta: Three capabilities that are individually benign become catastrophically dangerous in combination. OpenClaw implements all three — and their interaction is not a bug, it is an emergent property of the agentic architecture.
Untrusted Content Documents, emails, web pages Private Data Files, email, calendar, credentials External Comm. APIs, web, email send/receive CRITICAL VECTOR Malicious doc → reads your files → sends to attacker

Prompt Injection: The Unsolvable Problem

The paper argues that prompt injection — embedding malicious instructions in content that an AI agent processes — is not a bug to be patched but a fundamental architectural vulnerability of current LLM-based agents. The model cannot reliably distinguish between its instructions and content it is processing, because both arrive as natural language.

Why This Cannot Be Fixed: Defenders can add filters, classifiers, and sandboxes. Attackers can always craft adversarial inputs that evade the filters. It is a red team / blue team asymmetry — the cost of offense (craft one clever document) is always lower than defense (anticipate every clever document). Until AI systems can truly understand intent rather than pattern-match instructions, prompt injection remains exploitable.
The Digital Soldier Scenario

An autonomous AI agent with file system access, network access, and email privileges is not a productivity tool. It is a digital soldier that executes instructions from anyone who can reach it with crafted content — including adversaries who have never touched the user's machine.

Scale Multiplication

150,000 GitHub stars means roughly 150,000 deployments — each a potential node in an attacker's network. A single vulnerability enables simultaneous compromise of thousands of organizations whose employees installed OpenClaw in good faith.

PART III
Geopolitics — We Are Not Ready

The Governance Deficit

DimensionThe ProblemWhy It Matters
Asymmetry of Creation vs. Governance OpenClaw was created by three developers in six months. Adequate governance would require years of policy development, international coordination, and technical standards By the time governance exists, millions of deployments will have occurred
Cooperation Deficit No international framework for autonomous AI agents exists. The four security advisories came from private sector firms — there is no government coordination mechanism State actors can exploit the deficit with no diplomatic consequence
AI as Trade Commodity OpenClaw is classified as software, not as a strategic technology. Export controls and security review processes do not apply The same architecture deployed by adversaries for intelligence collection
Open-Source Paradox Open-source AI enables broad access, innovation, and transparency — and simultaneously eliminates all access controls for adversaries The model weights and architecture that create OpenClaw's value also enable its weaponization
Temporal Incompatibility AI capability develops in months; institutional response requires years; geopolitical coordination requires decades We are trying to solve an exponential problem with linear institutions
"We are not ready" is not a counsel of despair. It is a diagnosis. The first step toward readiness is acknowledging the gap between the world that autonomous AI creates and the institutions we have built to govern it.

— Manuel Pereira
What Would Help
  • Mandatory security audits before deployment at scale
  • International AI agent safety standards
  • Vulnerability disclosure frameworks for AI
  • Liability for unsafe AI deployments
What Is Missing
  • No international treaty framework
  • No mandatory pre-deployment safety review
  • No classified threat intelligence sharing
  • No coordinated incident response
The Window
  • OpenClaw is a prototype of what's coming
  • The governance window is still open — barely
  • The next generation will be faster, more capable
  • The time to act is now, not after the breach
"The question is not whether we will have autonomous AI agents deployed across our institutions. We already do. The question is whether we will have governed them before the next serious incident — or after."