OpenLeash Adds a Human Check to Risky AI Agent Actions

Max Brin is developing OpenLeash, a product described as an ‘AV for AI’ intended to mitigate the risks associated with increasingly autonomous AI agents. The core functionality of OpenLeash is to add a human-in-the-loop authorization layer, acting as a guardian angel to prevent AI agents from causing unintended harm.

Agents tend to inherit the permissions of their user, but lack human situational awareness, leading to potentially wide access and generous permissions across a network. A single misconstrued command, malicious tool, or compromised model can cause significant damage. OpenLeash intercepts agent intentions and, based on user-defined configurations, prompts the user to approve actions that might be risky.

Users can specify acceptable API endpoints, destinations, or payment limits. For example, payments below a certain threshold could be automatically approved, while those exceeding that limit would require human authorization. The configuration is highly flexible and can be adjusted at any time.

Currently, OpenLeash is in active use by hundreds of personal users and at least four organizations. Brin highlights its relevance in a new era where AI empowers non-technical users to create agents, increasing the potential for unchecked agentic behavior. He notes that the product is designed to address the growing need for AI defense, particularly as AI enables individuals with no coding experience to build and deploy AI agents.

Related: UK Government Rolls Out Agentic AI Defense Plan Alongside Industry Pledge

Related: Critical Vulnerability Exposes GitHub Agentic Workflows to Prompt Injection

Related: Agentic AI Security: Wrong Context, Wrong Decisions at Machine Speed