Last updated:
The process of verifying that an AI agent's planned sub-actions remain consistent with the original human-provided objective.
In a multi-step reasoning chain, an agent can experience Goal Drift. Intent validation acts as a secondary check-sum that evaluates each intermediate step against the high-level mission. If the agent's plan involves an action that doesn't align with the original goal (e.g., trying to access payroll data while searching for a flight), the action is blocked.
Real world example:
A procurement agent plans to buy 50 laptops. Before the purchase API is called, the Intent Validation layer confirms that Buying Laptops matches the original human instruction to Replenish IT hardware.
