Experts Warn of OpenClaw AI Risks
Security risks associated with emerging AI platforms like OpenClaw are a growing concern for businesses. On the *This Week in Startups* podcast, panelists warned against installing such tools on work computers, citing risks of exposing sensitive information and passwords. The consensus is that widespread enterprise adoption will depend on developing better security guardrails and clearer regulatory frameworks.
- Recent research from SecurityScorecard's STRIKE Threat Intelligence team revealed tens of thousands of OpenClaw instances exposed to the public internet, many vulnerable to Remote Code Execution (RCE), which allows an attacker to run any code on the system. - A critical vulnerability, CVE-2026-25253, allowed for a complete compromise of the OpenClaw gateway with a single click on a malicious link, giving an attacker full administrative control. This and other command injection vulnerabilities were later patched. - Previously known as "Clawdbot" or "Moltbot," OpenClaw's design is considered inherently risky because it combines privileged access to a user's sensitive data and accounts with the ability to browse the web and interact with untrusted sources like emails and chat messages. - The Open Web Application Security Project (OWASP) lists prompt injection and training data poisoning as two of the most critical security threats for Large Language Models (LLMs). Prompt injection can manipulate the AI to bypass security, generate malicious content, or disclose sensitive information. - The risk extends beyond direct attacks; if an employee connects an AI agent to corporate systems like Slack or SharePoint, the agent's tendency to store unencrypted secrets and tokens in one place creates a significant risk for a deep network compromise. - Enterprise adoption of AI is rapidly increasing, with one 2025 survey indicating that 62.5% of companies in the information and communication sector are already using AI. However, a lack of trust in AI vendor security is a major concern for 38% of enterprise leaders. - Attackers can also attempt "model theft," which involves stealing the proprietary LLM model itself. This can lead to the loss of a company's competitive advantage and any sensitive information contained within the model. - While AI presents new security risks, it is also being leveraged for defense. Security teams are increasingly using AI to predict and identify threats more efficiently, automate responses, and analyze vast amounts of data for vulnerabilities.
