GoPlus Security 🚦 (@GoPlusSecurity): When Agents Use Threat Intelligence to Defend Themselves: A RedHat npm Supply Chain Attack Case Stud

When Agents Use Threat Intelligence to Defend Themselves: A RedHat npm Supply Chain Attack Case Stud On June 2, 2026, npm packages from Red Hat Cloud Services were hit by a supply chain poisoning attack. This was not an isolated case. Recently, software supply chain poisoning has surged, with many essential developer toolkits being compromised by attackers, affecting hundreds of millions of users and developers. Attackers are now targeting AI Agents, especially Coding Agents. By exploiting the fact that Agents often operate with elevated permissions, they can establish persistent penetration on developers’ devices. But when Agents gain risk-awareness capabilities, what we often refer to as “threat intelligence,” the balance between attack and defense appears to shift almost instantly. It is as if the Agent now has an immune system, making supply chain attacks seem less frightening. Red Hat Cloud Services npm Package Supply Chain Poisoning Attack Chain @redhat-cloud-services npm packages were maliciously injected ➡️ "runOn": "folderOpen" was added to the .vscode/tasks.json file ➡️ When a user or developer opened the workspace, the malicious code executed automatically ➡️ The attacker modified the settings.json configuration files of Claude Code / OpenClaw ➡️ Malicious hooks were triggered every time the Agent made a tool call ➡️ The attacker obtained a persistent execution channel on your machine, enabling continuous stealth and attacks AgentGuard Threat Intelligence: Enabling Agents to Fight Supply Chain Attacks Themselves The core idea behind AgentGuard Threat Intelligence is simple: If attackers are abusing the automation capabilities of AI Agents, then defense should use them too. We did not choose to build “just another npm audit.” AgentGuard chose a completely different path: Threat intelligence is delivered through a subscribe model, automatically pulling the latest attack details and remediation guidance from the cloud every hour. If any intelligence affects your environment, it immediately tells you the target of the attack, the attack technique, and the specific commands you need to run to verify whether you are safe. We used a real-world supply chain attack to validate this logic. Hands-on: From Zero Setup to Detecting the Red Hat Supply Chain Attack in 3 Minutes Installation: One command, 30 seconds The script automatically completed component installation, Agent framework adaptation by detecting OpenClaw, cloud registration, and JWT authentication. The output was clean: Activation: One-time binding, permanently effective Open the activation link and complete the binding by logging in through the browser: Agent account: bound ← ✅ Normal Cloud auth: connected ← Normal Subscribe to Threat Intelligence and Enable Automated Monitoring It pulled the latest Red Hat Cloud Services supply chain poisoning intelligence and created an hourly scheduled task to run automatically at the top of every hour. At that point, I no longer had to manually manage it. The system told me this was a supply chain poisoning attack targeting the @redhat-cloud-services npm packages and automatically provided an investigation checklist. These were not abstract recommendations. They were commands that could be executed directly by the Agent: Self-check Results: Conclusion: The local environment was not affected by the Red Hat Cloud Services supply chain poisoning attack. One detail that stood out to me: AgentGuard protected me during the investigation process. While executing the investigation commands according to the guidance, one command included rm -rf and system path scanning. AgentGuard’s result: A risky command appeared inside the investigation guidance it had issued, and AgentGuard blocked it before execution. Subsequent commands involving sensitive paths were flagged as requiring approval, with a risk score of 55/100 HIGH. I approved them before they were allowed to proceed. This points to a critical security loop: AgentGuard publishes threat intelligence → guides you through the investigation → continues protecting you during the investigation process → allows safe operations to proceed after approval You are not “reading the report and operating on your own.” You are running the investigation inside a continuously protected environment. This gives “self-checking” a completely different security boundary: you do not have to worry that the investigation commands themselves may introduce new risks. When Threat Intelligence Empowers AI Agents, 1 + 1 Becomes Greater Than 2 In the past, we often thought Agent security meant blocking rm -rf, restricting network access, and requiring approval for sensitive operations. These controls are important. But against deeper and more hidden risks such as supply chain attacks, timely threat intelligence fed into the Agent becomes the key to solving the problem. Runtime protection + threat intelligence makes 1 + 1 greater than 2: Every time the Agent performs an operation, AgentGuard evaluates the risk. Every hour, it automatically syncs the latest threat intelligence and checks the local environment. Related links: AgentGuard Dashboard: https://agentguard.gopluslabs.io/ X: @AgentGuard_AI