Assure – your AI agent says "done." This checks if it's lying (github.com)

The newly announced Assure is an innovative tool that enables AI agents to operate within a controlled environment to verify their claims, ensuring accountability and trustworthiness. Developed to run on a disposable SQLite and file workspace, the tool routes all agent activities through a gateway that enforces policies and relies on deterministic human-written assertions for validation. This design negates the need for external API keys or large language models (LLMs), emphasizing a self-contained verification process for the agent’s output. This development is significant for the AI/ML community as it addresses the critical issue of trust in AI systems by providing a mechanism to audit AI-generated claims and ensure compliance with expected behavior. Assure allows for detailed assessments of agents across various scenarios, identifying failures like false successes or unauthorized changes, leading to a more robust framework for evaluating AI behavior. With its ability to benchmark different models against each other, it promotes transparency and can catalyze advancements in trustworthy AI development, making it a vital resource for researchers and practitioners aiming to enhance the reliability of AI systems in practical applications.