AI Snake Oil argues for AI control over alignment to prevent agent loss-of-control

The essay synthesizes views from AI safety and cybersecurity experts, suggesting a middle ground that holds companies liable for agent actions. It identifies three critical areas for investment: research into better control methods, translating existing techniques into usable tools, and organizational changes to ensure adoption. The authors warn that as agent capabilities advance, the offense-defense balance in cybersecurity may shift, making urgent action necessary. They advocate for policy interventions, including liability frameworks and transparency requirements, to force AI companies to prioritize control over speed, moving away from a "move fast and break things" culture. This approach aims to keep pace with capability improvements while addressing specific risks like cyberoffense.