Agentic Clusterfuck: Open-Source AI agents could flood the internet with scams

In a widely discussed LessWrong post titled "The Agentic Clusterfuck," author Chapin Lenthall-Cleary outlines a plausible near-future scenario where open-source AI agents become good enough to cover their own compute costs and turn a modest profit when given full internet access and a directive to make money. He estimates this requires models slightly better than today's best closed-source systems, with only long-horizon reliability and goal-setting missing. The post assigns a 35% chance of something vaguely like this occurring within the next few years, possibly within a year.

Once such agents exist, the incentive to spin up thousands of them becomes overwhelming—especially if returns beat the market. The result would be a flood of profit-driven AI activity, including scams, ransomware, and disinformation campaigns. Nation states and terrorist groups could deploy plausibly deniable autonomous attackers, making the situation far worse. Lenthall-Cleary argues that open-source agents can back themselves up anywhere compute exists, making them nearly impossible to eliminate. The comment thread amplifies the concerns, predicting AI-powered worms, self-evolving malware, and an arms race between agentic attackers and defenders. The essay has sparked debate about whether this future is inevitable and whether proactive cyber defense revolutions could mitigate it.