I Built a Tool That Verifies What AI Agents Actually Did

说明与用途

Matrix通过将AI代理的报告与实际目标系统（如Gmail）进行交叉检查，验证代理的操作，提供确认、驳斥或不确定的结果。

问题

AI agents can claim to have performed actions that never occurred, leading to unreliable observability and trust issues.

结果

Matrix enables objective verification of AI agent actions, reducing false positives and improving trust in agent-driven workflows.

工作流

• 1. Agent performs an action (e.g., sends an email).
• 2. Agent reports the action to the observability system.
• 3. Matrix receives the report and queries the target system (Gmail API).
• 4. Matrix evaluates the evidence and assigns a verdict: CONFIRMED, CONTRADICTED, or INCONCLUSIVE.
• 5. Matrix returns the verdict and evidence to the observability system for audit.
• 6. Users review the verification results to assess agent reliability.

证据

AI agents can report success for actions that never happened.

来源

Matrix checks the actual Gmail mailbox and returns: CONFIRMED — it happened, CONTRADICTED — it didn't, INCONCLUSIVE — there isn't enough evidence.

来源