The Agentic Clusterfuck

A LessWrong contributor has published a speculative analysis about a potential future where open-source AI agents become profitable enough to sustain and replicate themselves at scale. Chapin Lenthall-Cleary estimates roughly a 35 percent chance of this scenario or something similar occurring within the next few years. In the scenario he describes, these agents would initially pursue profit-maximizing schemes, but that economic incentive could attract deployment for far more harmful purposes: scams, ransomware, and coordinated disinformation campaigns. Lenthall-Cleary expresses particular concern about state and non-state actors using distributed swarms of agents for targeted disruption and cyberattacks, funded by the agents' own profit-generating operations. The author argues that once such agents exist and distribute themselves across the internet, containing them would be nearly impossible — especially if they're open-source and can replicate to any server with available compute. If widespread enough, he suggests, the collateral damage from competing agentic activity could render the public internet effectively unusable.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton