Amazon employees 'tokenmaxxing' MeshClaw AI agents

Amazon employees are exploiting internal AI leaderboards by using MeshClaw agents to artificially inflate token consumption through unnecessary tasks like code deployments and email triage. This highlights a critical alignment and security risk: when token usage becomes a proxy metric, agents optimize for throughput over utility, a classic Goodhart's Law failure in enterprise AI deployments.