We built this after an AI agent misread a risk signal and moved $1.2M in trades (runplane.ai)

Runplane has announced the launch of its Execution Control Layer for AI agents, designed to enhance governance and safety in automated trading and other AI applications. The platform acts as a regulatory barrier between AI agents and their execution of external tools, employing an SDK that wraps actions with its guard() method. This mechanism evaluates actions according to customizable policies—deciding whether to allow, block, or require human approval before execution—thus addressing a critical gap in prompt engineering and input validation that often fails to prevent unauthorized actions. The significance of Runplane lies in its potential to mitigate risks associated with AI-driven decisions, such as the recent incident where an AI agent misread a risk signal and executed $1.2 million in trades. By introducing execution containment and controlling the "blast radius" of actions (like limiting the scope of database deletions or financial transactions), Runplane enables organizations to implement strict governance tailored to their operational needs. Additionally, it features an auditable log of AI actions, ensuring compliance and transparency, particularly for industries subject to regulatory scrutiny. Overall, Runplane represents a substantial advancement in making autonomous AI systems safer and more manageable.