Adversa AI, the global leader in Agentic AI Security and Continuous AI Red Teaming, today announced the release of SecureClaw, an open-source security suite designed specifically for the OpenClaw agentic AI platform. The solution provides a robust defense layer, auditing installations for vulnerabilities and equipping agents with behavioral rules to resist manipulation.
SecureClaw addresses critical security issues by combining a plugin with a cognitive skill. The solution covers the entire OWASP® Foundation Top 10 for Agentic AI, mitigating risks ranging from tool misuse and privilege abuse to goal hijacking.
Comprehensive auditing: The plugin performs over 51 automated audit checks, scanning for misconfigurations, exposed gateway ports, and weak file permissions.
Automated hardening: Includes 5 hardening modules to instantly secure the agent’s environment, such as binding gateways to localhost and locking down sensitive directories.
Behavioral defense: The skill loads 12 lightweight behavioral rules (approximately 1,150 tokens) into the agent’s context, guiding it to reject suspicious instructions and require approval for high-risk actions.
Supply chain protection: Scans for “ClawHavoc” malware signatures and similar malicious patterns in third-party skills.
Continuous monitoring: Runs 3 background monitors to detect file tampering and unauthorized credential access in real-time.
“SecureClaw is a small but meaningful open-source contribution aimed at quickly improving security in the young OpenClaw ecosystem. Many of these agents are run by enthusiasts with zero security in their setups, allowing anyone to extract API keys and other valuable information with a simple crafted message.
The protection isn’t perfect — no one in the industry can fully prevent the fundamental issue of prompt injections. However, it significantly raises the bar for attackers by blocking known attack patterns at the code level before they execute. Installation takes about 30 seconds, a very modest requirement for OpenClaw users,”
said Alex Polyakov, CTO and co-founder of Adversa AI
OpenClaw has emerged as a popular platform for autonomous AI agents, granting LLMs the ability to execute complex tasks by accessing local files, managing credentials, and interacting with external APIs and networks. However, this autonomy introduced significant risks. Agents with access to sensitive files and the internet are vulnerable to prompt injection, credential theft, and supply chain attacks. Without proper safeguards, an agent can be tricked into exfiltrating private data or executing malicious commands through its toolset.
Adversa AI protects enterprises by continuously stress-testing GenAI applications, AI agents, and MCP-based architectures to identify and fix vulnerabilities before deployment. The company works with Fortune 500 companies, financial institutions, and AI Startups building next-generation AI systems.
Learn more about the Adversa AI Red Teaming Platform.
Stay up to date with what is happening! Plus, get a first look at news, noteworthy research, and the worst attacks on AI—delivered right to your inbox.
A detailed, practical guide to unintended code execution in agentic AI systems, covering risk profiles, attack vectors, and concrete mitigations for security teams and architects.
